[ { "id": 1073, "name": "Pathmind", "slug": "pathmind", "former_names": [ "Skymind" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/6f25b083ae5853d0fb4e5eee2ce6794565ac6775.png", "website": "https://pathmind.com", "all_locations": "San Francisco, CA, USA; Remote", "long_description": "Pathmind helps industrial engineers and simulation modelers achieve 10-20% gains in performance for industrial operations, supply chains and mining. We help them apply deep reinforcement learning to simulations and digital twins, and deploy decision-making policies into operations. ", "one_liner": "Pathmind helps industrial engineers and simulation modelers achieve…", "team_size": 15, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1447748416, "tags": [ "Reinforcement Learning", "Warehouse Management Tech", "Supply Chain" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2016", "status": "Inactive", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada", "Remote", "Fully Remote" ], "stage": "Growth", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/pathmind", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2016/pathmind.json", "year_founded": 2014, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/37560379/", "x_url": "https://twitter.com/pathmindAI", "fb_url": "https://www.facebook.com/PathMind-113337653394088/", "cb_url": "https://www.crunchbase.com/organization/skymind", "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/6f25b083ae5853d0fb4e5eee2ce6794565ac6775.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": null, "founders": [], "jobs": [], "news": [ { "title": "Deep reinforcement learning will transform manufacturing as we know it – TechCrunch", "url": "https://techcrunch.com/2021/06/17/deep-reinforcement-learning-will-transform-manufacturing-as-we-know-it/", "date": "Jun 17, 2021" }, { "title": "Supply Chain Optimization and the Future of Industry | Automation World", "url": "https://www.automationworld.com/business-intelligence/article/21485607/supply-chain-optimization-and-the-future-of-industry", "date": "Jun 17, 2021" } ], "launches": [] }, { "id": 12299, "name": "rct AI", "slug": "rct-studio", "former_names": [ "rct studio Inc.", "rct studio" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/c512748c3ff81d34f08d930a9b0d1edc1f08e8d5.png", "website": "http://rct.ai", "all_locations": "Los Angeles, CA, USA", "long_description": "rct AI is providing AI solutions to the game industry and building the true Metaverse with AI generated content. By using cutting-edge technologies, especially deep learning and reinforcement learning, rct AI creates a truly dynamic and intelligent user experience both on the consumers’ side and production’s side.\r\n\r\nThe founding team ever built a company, Raventech together and helped make it acquired by Baidu (NASDAQ:BIDU) in 2017. ", "one_liner": "build AI NPCs", "team_size": 40, "industry": "Consumer", "subindustry": "Consumer -> Content", "launched_at": 1544133609, "tags": [ "Reinforcement Learning", "Gaming", "Metaverse" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2019", "status": "Active", "industries": [ "Consumer", "Content" ], "regions": [ "United States of America", "America / Canada", "Remote", "Partly Remote" ], "stage": "Growth", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/rct-studio", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2019/rct-studio.json", "year_founded": 2018, "city": "Los Angeles", "country": "US", "linkedin_url": "https://www.linkedin.com/company/rct-ai", "x_url": "https://twitter.com/rct_ai", "fb_url": null, "cb_url": null, "github_url": "https://github.com/rct-ai", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/c512748c3ff81d34f08d930a9b0d1edc1f08e8d5.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": null, "founders": [ { "user_id": 306996, "full_name": "Xinjie Ma", "title": "Founder", "founder_bio": "Xinjie Ma is a co-founder of rct studio. She jointly founded rct studio with other former members of the Raven Tech team, which was acquired by Baidu in 2017. \n\nFor her work designing consumer technology user interfaces and marketing, she was recognized among the 2020 Forbes 30 Under 30 for both Asia and China.\n\n", "is_active": false, "linkedin_url": "https://www.linkedin.com/in/xinjie-ma-06aa3a304/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/b3816997bdb069d8579c3c04e0df1fdd372524c8.jpg" }, { "user_id": 306995, "full_name": "Cheng Lyu", "title": "Founder", "founder_bio": null, "is_active": false, "linkedin_url": null, "x_url": null, "avatar_thumb_url": "/avatars/thumb/missing.png" }, { "user_id": 306997, "full_name": "Xiaolong Mu", "title": "Founder", "founder_bio": null, "is_active": false, "linkedin_url": null, "x_url": null, "avatar_thumb_url": "/avatars/thumb/missing.png" }, { "user_id": 274168, "full_name": "Xiankun Wu", "title": "Founder", "founder_bio": "", "is_active": false, "linkedin_url": "https://www.linkedin.com/in/xiankunwu", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/1d335b89dc0d0f10113b32cc8464aa401e68530f.jpg" }, { "user_id": 300609, "full_name": "Chen Yuheng", "title": "Founder/COO", "founder_bio": "", "is_active": false, "linkedin_url": "http://www.linkedin.com/in/sergiochan", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/15b15d15669ddb57900b00349e93dfe8eb7d350e.jpg" } ], "jobs": [], "news": [ { "title": "AI-powered AAA Web3 game Delysium nets US$10M strategic funding | e27", "url": "https://e27.co/ai-powered-aaa-web3-game-delysium-nets-us10m-strategic-funding-20221107/", "date": "Nov 07, 2022" }, { "title": "Delysium Raised $10 Million in a Strategic Round | by Delysium | Delysium | Oct, 2022 | Medium", "url": "https://medium.com/delysium/delysium-raised-10-million-in-a-strategic-round-58a68b0328c7", "date": "Oct 18, 2022" }, { "title": "Delysium Raised $10 Million in a Strategic Round – CryptoMode", "url": "https://cryptomode.com/delysium-raised-10-million-in-a-strategic-round/", "date": "Oct 17, 2022" }, { "title": "We Closed Our Series A-3 Funding", "url": "https://rct.ai/en-us/blog/we-closed-our-series-a-3-funding", "date": "Nov 09, 2021" } ], "launches": [] }, { "id": 26837, "name": "Unify", "slug": "unify", "former_names": [ "Deep Ivy Inc.", "Ivy", "Unify", "Ivy" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/90e213981ccee521dac2989633ba8959528e318e.png", "website": "https://unify.ai", "all_locations": "London, England, United Kingdom; Remote", "long_description": "Building the next frontier of models that continually learn from a stream of experience.", "one_liner": "Models that keep learning", "team_size": 3, "industry": "B2B", "subindustry": "B2B -> Productivity", "launched_at": 1657390616, "tags": [ "Reinforcement Learning", "Open Source", "Infrastructure", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2023", "status": "Active", "industries": [ "B2B", "Productivity" ], "regions": [ "United Kingdom", "Europe", "Remote", "Fully Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/unify", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2023/unify.json", "year_founded": 2023, "city": "London", "country": "GB", "linkedin_url": "https://www.linkedin.com/company/unifyai", "x_url": "https://twitter.com/letsunifyai", "fb_url": null, "cb_url": "https://www.crunchbase.com/organization/unify-ai", "github_url": "https://github.com/unifyai", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/90e213981ccee521dac2989633ba8959528e318e.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jared Friedman", "url": "https://www.ycombinator.com/people/jared-friedman" }, "founders": [ { "user_id": 949841, "full_name": "Daniel Lenton", "title": "Founder", "founder_bio": "Love to dive deep into engineering rabbit holes 🕳️🐇", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/daniellenton", "x_url": "https://twitter.com/DanielLenton1", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/7f9d4085e00b2a0b42cbb6845e6ad5a9a6d40bb8.jpg" } ], "jobs": [], "news": [ { "title": "In 2024, it really is better to run a startup in San Francisco, according to data and founders who’ve relocated | TechCrunch", "url": "https://techcrunch.com/2024/08/25/in-2024-it-really-is-better-to-run-a-startup-in-san-francisco-according-to-data-and-founders-whove-relocated/", "date": "Aug 25, 2024" }, { "title": "Unify helps developers find the best LLM for the job | TechCrunch", "url": "https://techcrunch.com/2024/05/22/unify-helps-developers-find-the-best-llm-for-the-job/", "date": "May 22, 2024" } ], "launches": [ { "id": 81027, "title": "Unify - The best LLM on every prompt ✨", "tagline": "Dynamically route each prompt to the best LLM, so you can balance quality, speed and cost with ease", "body": "**TL;DR** Unify dynamically routes each prompt to the best LLM and provider so you can balance quality, speed, and cost with ease ✨\n\n🔀 [Try out our router in the browser](https://unify.ai/chat?default=true)\n\n💸 [Sign up now and get $50 free credits!](https://console.unify.ai/)\n\n🧑‍💻 [Get a Demo](https://calendly.com/unify-chat/general)\n\n**💥Problem** \\\nThere is a new LLM emerging almost every week, and all LLM applications have unique requirements for output quality, cost, and speed. It’s also unclear which models provide the highest quality on your task. This results in lots of manual signups, testing different models, bespoke benchmarking, etc. It’s overwhelming, and the final solution is almost always sub-optimal.\n\n![](https://t9015538723.p.clickup-attachments.com/t9015538723/65677e25-4ec6-4c1a-b235-56ff1fee6552/splash3.png)\n\nMany people give up and just use the largest models for everything. However, you really don't need GPT4 to summarize simple documents. Llama 8B is more than capable and is **15 times faster** and **100 times cheaper**. LLM apps are currently _much_ slower and more expensive than needed 💵🔥⏱️, and also often underperform in output quality because of poorly selected models 📉 (Claude Opus vs. GPT4 vs. Gemni etc.) for your specific prompts.\n\n**🤩 Not Anymore**\n\nUnify automatically routes each prompt to the most suitable model based on your preferences for quality, speed, and cost. Simply tune these three dials and let Unify do all the hard work.\n\n![uploaded image](/media/?type=post&id=81027&key=user_uploads/949841/fa900ddd-6849-408f-9a6c-1ad5dcfb0872)\n\nYour \"easy\" prompts will go to the fastest and cheapest models, and only the \"hard\" prompts will go to the most appropriate heavy lifters, like GPT-4o, Opus, and Gemini. You focus on building your LLM app, and we'll focus on providing the best models, with the fastest providers, at the lowest cost ⚡\n\n![uploaded image](/media/?type=post&id=81027&key=user_uploads/949841/b59ee67a-6566-4728-b146-57b102cf3151)\n\nWatch our [explainer video](https://youtu.be/15wgxK1Cw0E?si=5G04-7tGRKPyxiNa) to learn more about the solution at a high level.\n\n**What Unify Offers ✨**\n\n* **⚙️ Control**: Choose which models and providers to route to and adjust three sliders: quality, cost, and latency.\n* 📈 **Self Improvement**: Unify automatically improves your LLM application over time as new models and providers emerge.\n* 📊 **Observability**: [Compare all models and providers](https://console.unify.ai/dashboard) and see which are best for your needs.\n* ⚖️ **Impartiality**: Unify treats all models and providers equally, ensuring unbiased quality, cost, and speed [benchmarks](https://console.unify.ai/dashboard).\n* 🔑 **Convenience**: With a single API key, [access all models and providers](https://unify.ai/docs/api/first_request.html) behind a single endpoint, queryable individually or via the router.\n* 🧑‍💻 **Focus**: Don't stress about updating models and providers; Unify handles it for you so you can focus on building great LLM products.\n\n**Getting Started 🛠️**\n\n[Sign up](https://console.unify.ai/), select your [router](https://unify.ai/chat?default=true), then `pip install unifyai` and you're ready to go!\n\n```\nfrom unify import Unify\nunify = Unify(\n api_key=\"UNIFY_KEY\",\n endpoint=\"router@q:1\",\n)\nresponse = unify.generate(user_prompt=\"Hello there\")\n```\n\n🙏 **Our Ask**\n\nGive Unify a try, and let us know what you think! [Sign up](https://console.unify.ai/), `pip install unifyai`, take our [router](https://unify.ai/chat?default=true) for a spin, and check out our [product walkthrough](https://youtu.be/ZpY6SIkBosE). If you're excited about Unify, tell a friend about it 😊 If you don’t think it’s useful, please tell us. We’d love to know!", "url": "https://www.ycombinator.com/launches/L4t-unify-the-best-llm-on-every-prompt", "votes": 48, "created_at": "2024-05-22T15:22:53.596Z" } ] }, { "id": 27882, "name": "Velos", "slug": "velos", "former_names": [ "GradientJ" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/1521d60e05f3fa5d010d16b7722a46d4fb586f90.png", "website": "https://www.tryvelos.com", "all_locations": "San Francisco, CA, USA", "long_description": "Velos helps insurance carriers and MGAs automate complex, manual back-office tasks with AI workers instead of overseas teams.\r\n\r\nWe take over workflows with the same responsibility of a traditional overseas Business Process Outsourcing (BPO) firm: mapping the customer’s existing process, creating the SOP, building the automation, and continuously improving the workflow after launch. \r\n\r\nUnlike other \"automation platforms\" we don't rely on our customers to build and manage the workflow, we take over it for them like a traditional outsourcing firm.\r\n\r\nWe're automating the repetitive work people hate to do.", "one_liner": "AI BPO for Specialty Insurance MGAs", "team_size": 3, "industry": "Fintech", "subindustry": "Fintech -> Insurance", "launched_at": 1674503496, "tags": [ "Reinforcement Learning", "Robotic Process Automation", "Insurance", "Automation", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2023", "status": "Active", "industries": [ "Fintech", "Insurance" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/velos", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2023/velos.json", "year_founded": 2023, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/tryvelos/", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/1521d60e05f3fa5d010d16b7722a46d4fb586f90.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Nicolas Dessaigne", "url": "https://www.ycombinator.com/people/nicolas-dessaigne" }, "founders": [], "jobs": [], "news": [], "launches": [ { "id": 68762, "title": "GradientJ - Build and manage GPT3 🤖 applications at scale", "tagline": "GradientJ helps product teams deploy large language models at scale. Companies use us to create GPT3 applications and monitor them in production.", "body": "### **What is it?**\n\nGradientJ is a web application to build, compare, and deploy prompts for large language models like GPT3.\n\n### **Is this for me?**\n\nGradientJ is for:\n\n* Product teams that are just **thinking of using something like GPT3** in their product and are **trying to figure out how to get started**.\n* Teams who are **already using LLMs** to delight their customers and are **trying to level up their workflow** with faster iteration, better organization, easy cost and performance comparisons, and the on-ramp to \\~_the mythical land of fine-tuning\\~_.\n\n### **What’s the Problem?**\n\nCurrently, the **road from exciting prototype output with ChatGPT to production-ready GPT3 deployment is unclear** and littered with potential pitfalls.\n\nEven for the teams that do get version 1 out the door, **the workflow often ends up as a stitched together hodgepodge of makeshift processes and tools.**\n\nLLM deployment is a leading cause of “am I doing this right?”-itis. Symptoms include:\n\n* A text document with examples copy-pasted into the OpenAI playground\n* A prompt comparison process that consists of an eyeball test followed by “looks better-ish”\n* Hours spent tweaking a prompt on one example only to find, in the end, it breaks on all the others\n* Nightmares of users getting nonsensical or offensive output from a hallucinating LLM\n\n### **How do we solve it?**\n\nWe made a prompt engineering interface that:\n\nAutomatically suggests starting points or improvements to your prompts.\n\n![uploaded image](/media/?type=post&id=68762&key=user_uploads/314229/aabfc294-5d59-4f02-a9a0-ff9ea4e2a1b0)\n\nTracks prompt versions and benchmark examples so you can always compare and deploy the best iteration.\n\n![uploaded image](/media/?type=post&id=68762&key=user_uploads/314229/2491d923-0562-4c55-aa1c-fe0837a3b899)\n\nDeploys to your own API endpoint with automatic cost and performance tracking in one click.\n\n![uploaded image](/media/?type=post&id=68762&key=user_uploads/314229/df784795-2290-4ce9-9fbe-90ec41fbb21b)\n\n# **The Ask:**\n\n* **Let us help:** If you’re trying to use LLMs like GPT3 in your application and want to discuss how to **get version 1 out the door** **FAST**, we’d love to talk.\n* **Try it out:** If you’re already using LLMs in your application, we’d love to talk.\n* **Spread the Word:** If anyone you know is using or thinking of using LLMs, we want to turn their ideas into reality.\n\nDrop your email here , email us at [founders@gradientj.com](mailto:founders@gradientj.com), or jump right in and book a time to talk [here](https://calendly.com/daniel-gradientj/meeting)!", "url": "https://www.ycombinator.com/launches/Ht4-gradientj-build-and-manage-gpt3-applications-at-scale", "votes": 67, "created_at": "2023-01-26T20:03:07.818Z" } ] }, { "id": 29347, "name": "Vibrant Labs", "slug": "vibrant-labs", "former_names": [ "Exploding Gradients", "Ragas" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/b03fadf62fa443c473d34c31148c9785ea34dceb.png", "website": "https://vibrantlabs.com/", "all_locations": "San Francisco, CA, USA; Remote", "long_description": "We work on methods to autonomously scaling evals/environments for post-training agents. ", "one_liner": "Autonomously scaling environments", "team_size": 6, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1709606147, "tags": [ "Developer Tools", "Generative AI", "Reinforcement Learning", "Open Source", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2024", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada", "Remote", "Fully Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/vibrant-labs", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2024/vibrant-labs.json", "year_founded": 2023, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/vibrant-labs-ai", "x_url": "https://x.com/vibrantlabsai", "fb_url": null, "cb_url": null, "github_url": "https://github.com/vibrantlabsai", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/b03fadf62fa443c473d34c31148c9785ea34dceb.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/5cae3dfa6fa895e2f4fa256800008c55151eb0eb.png" ], "primary_partner": { "name": "Nicolas Dessaigne", "url": "https://www.ycombinator.com/people/nicolas-dessaigne" }, "founders": [ { "user_id": 387451, "full_name": "Jithin James", "title": "Founder/CEO", "founder_bio": "Jithin James (jjmachan), believes that life's purpose is found in the toil in mastering a craft and the fulfilment one gets from using it for the greater good. For him, this means excelling in software development, particularly in open-source and AI tooling, to empower fellow developers and help materialize their ambitious visions into reality. His journey is a testament to skill, dedication, and a commitment to community enrichment.", "is_active": true, "linkedin_url": "https://linkedin.com/in/jjmachan", "x_url": "https://twitter.com/jjmachan", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/bcdfd4732c2b0bbf0b007a1d8cae49d902d64f71.jpg" }, { "user_id": 889503, "full_name": "Shahul ES", "title": "Founder/CTO", "founder_bio": "Shahul (aka ikka) discovered his fascination for AI in his sophomore year while pursuing his computer science degree. This early interest laid the foundation for his journey to becoming a Kaggle GrandMaster and led to his significant contributions to open-source AI, particularly in initiatives like OpenAssistant AI.", "is_active": true, "linkedin_url": "https://linkedin.com/in/shahules", "x_url": "https://twitter.com/Shahules786", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/5bdaa55aa2f773d9e6e61c8bbd6944d8661b5a80.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 79074, "title": "Ragas: Open-source evaluation and testing Infrastructure for LLM applications", "tagline": "Deploy your LLM application with confidence. ", "body": "### **TL;DR**\n\nWe are building [Ragas](https://www.ragas.io/) — an open-source evaluation and testing infrastructure for LLM application developers to deploy their applications in production with confidence.\n\n### About us\n\n![uploaded image](/media/?type=post&id=79074&key=user_uploads/889503/21a941e3-4798-4628-95cc-15332facb47e)\n\nWe’re [Jithin](https://linkedin.com/in/jjmachan) and [Shahul](https://linkedin.com/in/shahules)! Having met in college, we’ve collaborated on various projects for almost a decade now.\n\nJithin takes care of building the software and infrastructure. He was an early employee at Bento ML, where he built and maintained tools like [Bentoctl](https://github.com/bentoml/bentoctl), [Bentoml](https://github.com/bentoml/BentoML), and [Yatai](https://github.com/bentoml/Yatai). Shahul is responsible for AI research and engineering. He is a [Kaggle Grandmaster](https://www.kaggle.com/shahules) and a lead contributor to different open-source AI projects, including [Open-Assistant AI.](https://github.com/laion-ai/open-assistant)\n\n### Problem\n\nBefore 2023 software used to be written in code but with the emergence of foundational models software and applications are going to be compound systems containing code, prompts, and other components. This introduces several new problems\n\n* How do you select the best model or component suitable for your application from the abundance of available resources?\n* How do you test these systems and ensure continuous quality?\n* How do you derive insights from production to measure and improve your system?\n\nAs early adopters of this technology to build applications, we faced this problem while we were building RAG systems early last year.\n\n### Solution\n\nWe at [Ragas](https://docs.ragas.io/) make use of model-graded evaluations and testing techniques to ensure quality. This includes automated synthesis of test data points, explainable metrics, and adversarial testing.\n\nWe started by building this for RAGs, which is the most popular application of LLM as of today. Ragas is now the default open-source standard for evaluating RAG applications, processing over 4.7 million responses last month and used by engineers from enterprises like AWS, Microsoft, Databricks, Moody’s, UHG, and Tencent.\n\n![uploaded image](/media/?type=post&id=79074&key=user_uploads/889503/4da2d427-8bfe-4e4b-a76e-01c7a28ce747)\n\n### Our Ask\n\n* Checkout ragas on [GitHub](https://github.com/explodinggradients/ragas)\n* If you’re building RAG applications, consider [applying](https://forms.gle/TnqUHuY176m2juqL9) for the Ragas office hours program", "url": "https://www.ycombinator.com/launches/KZO-ragas-open-source-evaluation-and-testing-infrastructure-for-llm-applications", "votes": 94, "created_at": "2024-03-05T18:38:36.494Z" } ] }, { "id": 29353, "name": "Resonance", "slug": "resonance", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/f32584f82882e696b3984dfcf4bd700cce69fd5f.png", "website": "https://www.useresonance.com/", "all_locations": "San Francisco, CA, USA", "long_description": "Resonance hyper-personalizes MarTech campaign content and automatically refreshes and stores high performing content for re-use.", "one_liner": "Self-improving content bank for marketing", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Marketing", "launched_at": 1709332105, "tags": [ "Reinforcement Learning", "SaaS", "Subscriptions", "Marketing", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2024", "status": "Acquired", "industries": [ "B2B", "Marketing" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/resonance", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2024/resonance.json", "year_founded": 2023, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/99332802/admin/feed/posts/", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/f32584f82882e696b3984dfcf4bd700cce69fd5f.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Harj Taggar", "url": "https://www.ycombinator.com/people/harj-taggar" }, "founders": [ { "user_id": 570574, "full_name": "Jonathan Ou", "title": "Founder/CEO", "founder_bio": "Jonathan is the CEO of Resonance. Before Resonance, he worked at DoorDash on a voice AI to automatically place customer orders over the phone. Prior to DoorDash, he worked at Dropbox to help build an internal in-product messaging platform that generated $100 million ARR. Prior to Dropbox, he worked at Microsoft helping build a platform that enabled doctors and patients to meet virtually using Microsoft Teams.", "is_active": true, "linkedin_url": "https://linkedin.com/in/jto98", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e6a93656f11045a21c1ba0ab1bac3a79a69bcd6a.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 79089, "title": "Resonance - Optimally upsell users within your product", "tagline": "Effortlessly boost net revenue retention in your product post user onboarding.", "body": "> **TL;DR:**\n\n[_Resonance_](https://www.useresonance.com/) _is an ad server for house ads (self-promotional marketing). Companies use Resonance as a one-stop shop for campaign creation, intelligent experimentation/prioritization, analytics, and impression management to boost net revenue retention following user onboarding._ [_Schedule a demo_](https://calendly.com/jonathantou/quick-chat) or [_get in contact_](https://www.useresonance.com/contact).\n\n> **The Team:**\n\nI used to work as a SWE at Dropbox, DoorDash, and Microsoft. In particular, I helped build a similar platform to Resonance at Dropbox that generated $100 million ARR. I wanted to start Resonance because 1) I felt there were many ways to improve that particular tool and 2) I wanted to bring that tool to other companies that need it but don’t have the bandwidth to build it themselves.\n\n> **The Problem:**\n\nFor companies that depend on product-led growth (primarily consumer/SMB/startup-centric companies), their website remains their most valuable marketing channel. Post-user onboarding, these companies rely on in-product messaging to drive feature adoption so they can retain users.\n\nUnfortunately, this sort of strategy only works at scale, and there are many challenges in doing so. Below are some non-trivial functionalities that modern companies require to launch effective campaigns with velocity:\n\n1. Code-based and no-code workflows for launching campaigns\n2. Campaign performance analytics\n3. Intelligent campaign targeting and experimentation\n4. Sophisticated campaign impression management\n5. UX flexibility (i.e. messaging should not be limited just to overlays)\n\nSolving this problem is crucial, as millions in ARR is left on the table if companies cannot properly scale their in-product feature discovery/awareness/education. Companies like Dropbox that have built tooling to address the above challenges have generated up to $100 million ARR from the increased user retention.\n\n> **The Solution:**\n\nSolving this problem requires building a centralized system where any stakeholder (technical or non-technical) can effortlessly launch campaigns to drive in-product feature adoption.\n\nHolistic assessment of campaign performance, global impression management, and optimized campaign targeting can only work at scale if there is a single system that serves as a repository for all campaigns and governs campaign creation.\n\nResonance addresses this by providing a central platform where developers and non-developers can both easily launch in-product feature adoption campaigns within minutes, with the above functionalities provided out of the box.\n\nBelow are screenshots of some important features in our UI (and there are many more).\n\n* No code builder with live preview on your website and experimentation\n * (**NOTE**: The preview will obviously be with your website and UI components - disregard my modal and website)\n\n![uploaded image](/media/?type=post&id=79089&key=user_uploads/570574/7339d733-3683-43bd-9fac-f23d881bfdf4)\n\n* Complex User Targeting\n\n ![uploaded image](/media/?type=post&id=79089&key=user_uploads/570574/cc3cd9e1-809f-499e-b871-934e27807a38)\n\n* Sophisticated Frequency Management\n * (**NOTE**: We also have similar rules that apply globally for all campaigns)\n\n![uploaded image](/media/?type=post&id=79089&key=user_uploads/570574/f1b8a942-dc12-440f-8680-6c4dee517be4)\n\n* Global Analytics\n\n![uploaded image](/media/?type=post&id=79089&key=user_uploads/570574/62457d65-c845-48ff-8877-569dfa707544)\n\n* Intelligent Experimentation (experiment up to 7x faster with 4-8x more variants)\n\n ![uploaded image](/media/?type=post&id=79089&key=user_uploads/570574/53430985-a9d3-4c17-97eb-52bc56285c86)\n\n\n> Our Ask\n\n1. If you are a product-led company, shipping a lot of features, or are using something like Intercom or Pendo, I’d love to set up some time here to chat about improving your user retention. I’m familiar with existing tools and I believe I can do a much better job.\n2. If not, I’d appreciate an intro to anyone in your network who works in a senior role within growth or product if possible.", "url": "https://www.ycombinator.com/launches/KZd-resonance-optimally-upsell-users-within-your-product", "votes": 48, "created_at": "2024-03-06T01:22:47.987Z" } ] }, { "id": 29407, "name": "JustAI", "slug": "justai", "former_names": [ "Choice", "Choice AI", "just words", "Just Words" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/166536f94d496e2b50084b3b8edf8b6d9e29040b.png", "website": "https://getjust.ai", "all_locations": "San Francisco, CA, USA", "long_description": "Always-on AI agents for 1-1 personalization at scale", "one_liner": "Agentic Marketing Platform", "team_size": 4, "industry": "B2B", "subindustry": "B2B -> Marketing", "launched_at": 1704840202, "tags": [ "Reinforcement Learning", "Workflow Automation", "Personalization", "Marketing", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Winter 2024", "status": "Active", "industries": [ "B2B", "Marketing" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": true, "app_answers": null, "question_answers": [ { "question": "How did your company get started? (i.e., How did the founders meet? How did you come up with the idea? How did you decide to be a founder?)", "answer": "When we were heading Growth at Twitter, we saw the power of copy changes. **When we tweaked a push notification from “Elon tweeted” to “Elon _just_ tweeted”, we gained 800K additional users.** We never underestimated the power of copy since that day. The company funded it as a big investment in the difficult years when Twitter’s growth was tanking. However, each change would take 3-6 months to pick, test, approve, and ship. We also didn’t know how to use those learnings systematically to prioritize the next copy change. We started building internal tools to automate this and gained 20% additional user growth. We stumbled upon similar problems and “vision docs” at Pinterest. We knew we were staring at money on the table. We quit our jobs to build [just words](https://justwords.ai)." }, { "question": "What is your long-term vision? If you truly succeed, what will be different about the world?", "answer": "Today, lifecycle marketing is built on manual effort — copywriters brainstorming variants, data scientists running A/B tests, and marketers trying to make sense of disconnected dashboards. Every campaign is a one-off experiment.\r\n\r\nIn the world we’re building, **AI-marketers** will handle this entire loop — strategy, creation, optimization, and learning — with humans acting as orchestrators, not operators.\r\n\r\n* Campaigns will **self-evolve**: AI agents will detect new opportunities, write and test messages, measure outcomes, and continuously improve.\r\n* Brands will move from **batch campaigns** to **living systems** — always-on, hyper-personalized, cross-channel journeys that adapt in real time.\r\n* Every marketer, no matter their company size, will have access to enterprise-level intelligence — a strategist, a creative, an optimizer, and a data scientist — all in one AI team.\r\n\r\nThe world will shift from “humans using tools” to **humans teaming with AI agents**.\\\r\nAnd in doing so, we’ll make personalization not just scalable, but truly human again — every message relevant, every experience intelligent, every brand empathetic at scale." } ], "url": "https://www.ycombinator.com/companies/justai", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2024/justai.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/getjustai", "x_url": "https://twitter.com/getjustai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/166536f94d496e2b50084b3b8edf8b6d9e29040b.png", "app_video_url": null, "dday_video_url": "https://bookface-images.s3.us-west-2.amazonaws.com/video_event/64/1712293254075-oa1quvatg4-d244daa8dd4b243426185eb8f60afd01/97ff172a722b3330829f91f059c8a6fa.mp4?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIAQC4NIECABPQ2AHNW%2F20261020%2Fus-west-2%2Fs3%2Faws4_request&X-Amz-Date=20261020T022718Z&X-Amz-Expires=3600&X-Amz-SignedHeaders=host&X-Amz-Signature=9d901f8c8d5a89ffbc8593efcbecbf2e7710642a848ee8af318bba6194c23a5a", "company_photos": [], "primary_partner": { "name": "Pete Koomen", "url": "https://www.ycombinator.com/people/pete-koomen" }, "founders": [ { "user_id": 1381391, "full_name": "Neha Mittal", "title": "Founder/CEO", "founder_bio": "Neha Mittal is the CEO & Co-founder of JustAI. She was the head of product for Retention & Growth at Pinterest and at Twitter. She also co-founded CityStructure and was the founding engineer at Sensorflow. She has also held product & engineering roles at the Chan Zuckerberg Initiative and Goldman Sachs.", "is_active": true, "linkedin_url": "https://linkedin.com/in/neha01mittal", "x_url": "https://twitter.com/neha01mittal", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/163592976aa66718aaf2fdbb6a4f09a93ea4efb0.jpg" } ], "jobs": [ { "id": 89154, "title": "Full Stack Software Engineer", "url": "https://www.ycombinator.com/companies/justai/jobs/BmBq81w-full-stack-software-engineer", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$120K - $160K", "equity_range": "0.10% - 0.50%", "experience": "1+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 89155, "title": "Account Executive", "url": "https://www.ycombinator.com/companies/justai/jobs/NOV6eqU-account-executive", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$150K - $350K", "equity_range": "0.10% - 0.50%", "experience": "1+ years", "visa": "US citizen/visa only", "skills": [] } ], "news": [ { "title": "Just Words Gets Funding From Peak XV Partners, Y Combinator", "url": "https://inc42.com/buzz/former-x-executives-startup-just-words-nets-funding-to-automate-personalised-messaging-for-enterprises/", "date": "Aug 27, 2024" }, { "title": "YC’s latest Demo Day shows fascinating wagers on healthcare, chip design, AI and more | TechCrunch", "url": "https://techcrunch.com/2024/04/04/ycs-latest-demo-day-shows-fascinating-wagers-on-healthcare-chip-design-ai-and-more/", "date": "Apr 04, 2024" } ], "launches": [ { "id": 77393, "title": "just words - Scale Hyper-personalized Messaging with AI", "tagline": "Maximize growth velocity with automated message refresh on emails and push.", "body": "**tldr:** We’re building an AI-first marketing tech stack for enterprises. We are on a mission to rethink growth by turning content into conversations. Our core product allows companies to auto-refresh content on growth channels like emails using reinforcement learning, artificial intelligence, and causal inferences. Companies have 10X their speed of experimentation, and have seen stat sig engagement lift with our technology. \n\n```\n❌ the problem\n```\n\n1\\. Traditional A/B testing is slow. It can take 3-6 months to run one test because of organizational decision-making and lack of ML infrastructure. \n\n2\\. Content and copy selection is based on vibes. \n\n3\\. Prioritization of the biggest opportunities is decided by the loudest voice in the room. \n\n```\n✨ our solution\n```\n\n1\\. Enrich CRM with ML powered personalization \n\n2\\. Auto-refresh content on the fly. Auto-tune underperforming content with newly generated dynamic content in real time\n\n3\\. Use insights from infinite testing to predict the next best action.\n\n![uploaded image](/media/?type=post&id=77393&key=user_uploads/1381391/30b40cc4-6a73-4e12-9a09-5cbba4930fb3)\n\n```\n🙏 asks\n```\n\n1\\. Check out [justwords.ai](http://justwords.ai) and book a demo for your use case (email, sms, push, landing page titles).\n\n2\\. Connect us to your growth & marketing teams. Book a [30 min slot](https://calendly.com/just-words) with us or drop us an email at [founders@justwords.ai.](mailto:founders@justwords.ai)", "url": "https://www.ycombinator.com/launches/K8H-just-words-scale-hyper-personalized-messaging-with-ai", "votes": 70, "created_at": "2024-01-10T20:48:09.601Z" } ] }, { "id": 29825, "name": "Sepal AI", "slug": "sepal-ai", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/78c48b77fe0ad236d148f827dd2ba0b86450afd3.png", "website": "https://www.sepalai.com", "all_locations": "San Francisco, CA, USA", "long_description": "Sepal is a data research company on a mission to advance human knowledge and capabilities through safe AI.\r\n\r\nWe partner with the world’s leading AI labs and enterprises to help their models get better at the tasks people actually want them to do.\r\n\r\nWe’ve built a Cloud-Native Agent Dataset Factory which turns the process of generating evaluation and training data from manual, inconsistent, and labor-intensive into something automated, standardized, and scalable.\r\n\r\nSepal AI was founded in 2024 by engineers and operators from Vercel and Turing. We went through Y Combinator, raised several million dollars from leading investors, and already count multiple Fortune 500s and top AI research labs as paying customers.", "one_liner": "Data Development for Advanced AI", "team_size": 15, "industry": "B2B", "subindustry": "B2B", "launched_at": 1722966351, "tags": [ "AIOps", "Reinforcement Learning", "Data Labeling", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2024", "status": "Acquired", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada", "Remote", "Partly Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/sepal-ai", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2024/sepal-ai.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/sepalai/", "x_url": "https://x.com/sepal_ai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/78c48b77fe0ad236d148f827dd2ba0b86450afd3.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Tyler Bosmeny", "url": "https://www.ycombinator.com/people/tyler-bosmeny" }, "founders": [ { "user_id": 1996512, "full_name": "Robi Lin", "title": "Founder", "founder_bio": "Co-Founder @ Sepal AI\n\nBuilt the enterprise workflow products and fulfillment strategy at Turing.com. Scaled Turing’s LLM trainer business line from 50 to 800+ onboarded developers in 5 months for foundational LLM and enterprise customers. Previously was at Bain & Co.\n", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/robert-lin1/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/4668b119bdaddeda68e7781c88272985fc03575c.jpg" }, { "user_id": 860488, "full_name": "Kat Hu", "title": "Founder", "founder_bio": "Cofounder @ Sepal AI\n\nBuilt Turing’s Foundational LLM trainer business GTM. Ran orgs of 500+ AI trainers & built corresponding operations for scale. Previously was at McKinsey.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/kathleenhu/", "x_url": "https://www.twitter.com/katqhu1", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/0060040b5fd066406e4be3460a6c0d0209a59029.jpg" } ], "jobs": [], "news": [ { "title": "Sepal AI supports the launch of Anthropic's Claude 3.7 Sonnet with Pre-launch Testing", "url": "https://x.com/sepal_ai/status/1894157588142690695", "date": "May 01, 2025" } ], "launches": [ { "id": 82824, "title": "🌱 Sepal AI - Confidently deploy your AI models", "tagline": "We build benchmarks, evals, & training datasets for enterprises.", "body": "# **Tl;dr:** Sepal provides frontier data and tooling for advancing responsible AI development.\n\n# **__________________________________________________________**\n\n[Sepal AI](https://www.sepalai.com/?ref=launch_yc) is on a mission to advance human knowledge and capabilities with the responsible development of artificial intelligence.\n\n# **🧐 Responsibly advance human knowledge with AI? What does that mean?**\n\nWe believe in a world where AI advances scientific research and empowers economic growth. \n\nTo achieve that future, AI product & model builders need:\n\n1. **Golden Datasets and Frontier Benchmarking:** To iteratively measure model performance on specific use cases.\n2. **Training Data:** To improve model capabilities using fine-tuning and RLHF.\n3. **Safety / Red-teaming:** To measure and forecast the safety of LLMs before putting them out in the wild.\n\n# **__________________________________________________________**\n\n# ⚠️ **Okay, well why does it matter?**\n\nFrontier data for AI development is vital for safe deployment & scaling. However, developing this data is difficult. \n\nMost frontier data requires domain knowledge that can be hard to source and curate (e.g., finance, medical, physics, biology, etc.). Publicly available benchmarks (e.g., MMLU, GPQA, MATH, etc.) are contaminated and too general to be useful to actual product & model builders.\n\n# **__________________________________________________________**\n\n# **🌱 How do we do this?**\n\nWe’ve built Sepal AI - the data development platform that enables you to curate useful datasets.\n\n**The Platform:** We bring data generation tooling, human experts, synthetic data augmentation, and rigorous quality control into one platform so you can manage the production of high-quality datasets.\n\n![uploaded image](/media/?type=post&id=82824&key=user_uploads/860488/81dd4889-9694-4c67-8505-389edf7c3361)\n\n**Our Expert Network:** We’ve built a network of 20k+ experts across STEM and professional services (think academic PhDs, business analysts, medical professionals, marketing and finance consultants) to support campaign design & data development. \n\nSample engagements we’ve run:\n\n* **🧬 Cell and Molecular Biology Benchmark:** An original benchmark to evaluate complex reasoning across models. Produced by a team of PhD biologists from top institutions in the US.\n* **💼 Finance Q&A + SQL Eval:** A Golden Dataset to test the ability of an AI agent to query a database and produce human-expert-level answers to complex finance questions.\n* **📏 Uplift Trials & Human Baselining:** End to end support for conducting secure in-person evaluations on model performance. \n* _…. \\[insert your custom use case next?\\]_\n\n# **__________________________________________________________**\n\n# **🙏 Asks:**\n\n1. If you are building an AI application and need to measure or improve your model, or\n2. If you are a researcher at an AI lab building or evaluating models for new capabilities / risk areas, or\n3. If you’re passionate about the development of AI, AI safety, or evals in general…\n\nLet’s chat — please check us out at [www.sepalai.com](http://www.sepalai.com) \n\n# **__________________________________________________________**\n\n# **👪 Our team:**\n\n![uploaded image](/media/?type=post&id=82824&key=user_uploads/860488/a1187807-573a-4248-9181-6378d6e2b042)\n\nMeet [Kat](https://www.linkedin.com/in/kathleenhu/) (on the left), [Robi](https://www.linkedin.com/in/robert-lin1/) (in the middle), [Fedor](https://www.linkedin.com/in/f3d0r/) (on the right)!\n\nRobi and Kat previously built the technical LLM training business for Turing. Kat on the go-to-market & operations side. Robi on the product & fulfillment side. Fedor is a long-time close friend - he was an early engineer at Vercel & Newfront where he built out foundational infrastructure.\n\nSay hi: [founders@sepalai.com](mailto:founders@sepalai.com).", "url": "https://www.ycombinator.com/launches/LXs-sepal-ai-confidently-deploy-your-ai-models", "votes": 51, "created_at": "2024-08-07T18:46:33.791Z" } ] }, { "id": 30125, "name": "Synth", "slug": "synth-3", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/72c8ca1f80fbcb63a0a616d53bce31153601a954.png", "website": "https://www.usesynth.ai", "all_locations": "San Francisco, CA, USA", "long_description": "Choose a coding agent harness, model, and task dataset and optimize context and prompts to get the best performance for long-horizon tasks", "one_liner": "Prompt and Context Optimization for Coding Agents", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1729021221, "tags": [ "Reinforcement Learning", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Fall 2024", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/synth-3", "api": "https://devasheeshg.github.io/yc-api/batches/fall-2024/synth-3.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/105503800", "x_url": "https://x.com/UseSynth", "fb_url": null, "cb_url": null, "github_url": "https://github.com/synth-laboratories", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/72c8ca1f80fbcb63a0a616d53bce31153601a954.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Pete Koomen", "url": "https://www.ycombinator.com/people/pete-koomen" }, "founders": [ { "user_id": 2210691, "full_name": "Josh Purtell", "title": "Founder/CEO", "founder_bio": "Josh is the founder and CEO of Synth.\n\nBefore Synth, Josh was employee #3 and research scientist at Basis, which uses AI agents to automate accounting tasks. Previously, he built out the ML core of a Cyber Threat Intelligence SaaS product at Halborn, a cyber security startup that acqui-hired his startup Politeia. He founded Politeia while wrapping up a Math undergrad at Yale, where he did time-series ml research for fun.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/joshua-purtell/", "x_url": "https://x.com/JoshPurtell", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/a7308f11e9b3a936bc55aa7b591edf71b042a061.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 85819, "title": "Synth: Automated Training for Vertical AI Agents", "tagline": "Making developing and maintaining a SOTA AI system simpler.", "body": "Hi, I’m Josh, founder of Synth.\n\n**TL;DR**\n\nWe log what your agent is doing, identify where it’s coming up short, and fine-tune the AI models your agent uses to prevent those failures.\n\nYou add a few decorators to your code, run your agent, and get back fine-tuned models (via OpenAI or the provider of your choice) that perform better. Our product, when applied to SWE-Agent, an academic baseline, delivers a **33%** relative improvement on the [SWE-Bench agent benchmark](https://www.swebench.com) with just 100 training data points. [ ](https://www.youtube.com/watch?v=DwC8iBLpWdU&feature=youtu.be)\n\n**Why is this important?**\\\nTeams want to deploy software that automates economically meaningful tasks over multiple steps. Iterating on those systems is hard: an agent taking 100 steps to complete a task will often create a newspaper’s worth of text data over just one task run, and reviewing a representative evaluation dataset can be daunting. Moreover, getting language models to demonstrate context-specific agentic behavior - such as following strong plans and making the best use of their environment - can be challenging, just with manual prompt tweaking. \n\nWorking as an agent researcher at a startup automating accounting tasks, I struggled to parse through the reams of logs my systems would generate every time I wanted to test a potential improvement. Manually tweaking dozens of prompts and reasoning about how changes would propagate through the system was also difficult— and, for many problems, ineffective. As agents are deployed in more challenging applications and to more customers, these problems will grow. Agent developers need a tool that scales with their agent’s complexity and scope in order to build state-of-the-art systems.\n\n**Our solution**\\\nSynth works in 3 steps\n\n1. Communicate: We summarize your agent’s logs to help you understand how it’s performing at a glance.\n2. Flag: We identify common failure modes your agent is demonstrating.\n3. Address: We fine-tune base models your agent uses to address AI-related issues and suggest changes to your code when gaps in agent scaffolding are the root cause.\n\nWe worked with a startup that was working on automating code generation to increase their agent’s success rate at an editing task from ~**85% to 100%** on a representative evaluation set.\n\n**Why Now?**\\\nDevelopers and researchers have been building AI agents with language models for years, but only somewhat recently have base models provided a strong enough foundation to enable the most ambitious applications.\n\nMoreover, some of the most powerful and consistent approaches for tuning agents to address domain-specific challenges— the ones powering Synth— have only been published in the last year or so. The market is ready for our solution, as is the research literature.\n\n**Team**\\\nI recently worked at [Basis](https://www.getbasis.ai), a startup automating accounting tasks with AI agents, as a researcher developing and maintaining pipelines and agents used to serve top accounting firms. While there, I had the opportunity to publish [academic research](https://arxiv.org/abs/2406.11695) on the topic of agent optimization at EMNLP, a respected AI conference.\n\nBefore Basis, I built a startup doing ML for cyber with some friends that got acqui-hired right out of the gate for a modest amount. Before then, I studied math and machine learning at Yale.\n\n**Ask**\\\nSend folks building agents our way at [founders@usesynth.ai.](https://mailto:founders@synth.ai) We’d love to talk to anyone using AI agents to solve hard, multi-step tasks - the more ambitious and complex the approach, the better.\n\n![uploaded image](/media/?type=post&id=85819&key=user_uploads/89898/5d64c8db-48e1-423a-a020-f5ace9ea7b82)\n\n", "url": "https://www.ycombinator.com/launches/MKB-synth-automated-training-for-vertical-ai-agents", "votes": 82, "created_at": "2024-11-23T14:56:40.105Z" } ] }, { "id": 30173, "name": "Aviro", "slug": "aviro", "former_names": [ "Tropir" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/3625aa6dc8b9ca3fedb43ca1043c7246c8185306.png", "website": "https://www.aviro.ai/", "all_locations": "San Francisco, CA, USA", "long_description": "We design challenging worlds where AI models learn to explore and solve problems. We turn these worlds into post-training datasets and reinforcement learning environments to train the next frontier of AI. We're research partners with four top frontier AI labs, several Fortune 100 companies, and leading RL data vendors.", "one_liner": "Environments for Long Horizon Work", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1746601065, "tags": [ "Reinforcement Learning" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Spring 2025", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/aviro", "api": "https://devasheeshg.github.io/yc-api/batches/spring-2025/aviro.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/aviro-ai/", "x_url": "https://x.com/aviro_ai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/3625aa6dc8b9ca3fedb43ca1043c7246c8185306.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Garry Tan", "url": "https://www.ycombinator.com/people/garry-tan" }, "founders": [ { "user_id": 2318975, "full_name": "Aarush Kukreja", "title": "Founder", "founder_bio": "Building Aviro", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/aarush-kukreja/", "x_url": "https://twitter.com/aarush_kukreja", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/820110a893785c60a73ad43c2d4d46e8d028a95f.jpg" }, { "user_id": 2318974, "full_name": "Ayush Karupakula", "title": "Founder/COO", "founder_bio": "Building Aviro", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ayush-karupakula", "x_url": "https://x.com/ayushkarupakula", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/02430fba8423cfecce32f344331626743ff315ea.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 93860, "title": "Aviro - Continuous Upskilling for Enterprise AI Agents", "tagline": "Introducing Cortex, our runtime layer which gives agents the right insights at the right moments", "body": "**TL;DR:** Long-horizon agents break down. They can retrieve facts and utilize tools, but when it comes to long, multi-step workflows, they fail to reproduce successful behaviors and consistently repeat costly mistakes.\n\nAviro fixes that. We’re starting with Cortex, an RL-based intelligence layer that makes agents learn through experience. With Cortex running, our deep research agent outperformed OpenAI's—beating it by 70% on enterprise search tasks and reaching #1 on Microsoft's Deep Research benchmark.\n\n**We're partnering with teams building advanced agents for complex business. Contact us** at [founders@aviro.ai](mailto:founders@aviro.ai) or book a time [here](https://cal.com/aarush-kukreja/demo).\n\nhttps://youtu.be/8Sz16v4X63w\n\n## Why This Matters\n\nAgents are now being tasked with longer, higher-stakes enterprise processes, where mistakes can compound quickly, and reliability erodes rapidly. Today's agents:\n\n* **Don't carry forward what worked**\n* **Lose context across tools and handoffs**\n* **Repeat the same errors** (wasting time and compute)\n* **Deliver inconsistent results** teams can’t trust\n\nThey fail not because models aren't smart enough, but because agents can't learn or adapt to enterprise workflows.\n\n## What Makes Aviro Different\n\nOur intelligence layer, Cortex, solves this by providing task-time intelligence that evolves with your enterprise. Cortex:\n\n* **Learns From Every Attempt -** extracts reusable lessons from long, complex runs that help future ones stay on track\n* **Context-Aware Retrieval -** surfaces the right lessons at the right moments, without slowing or overloading the agent\n* **Workflow-Aware Guidance -** anchors to your **existing SOPs** while evolving them as better approaches emerge\n* **RL-Powered -** uses reinforcement learning to optimize guidance and retrieval over time\n* **Seamless Enterprise Integration -** works alongside your current agents, models, and tools with **lightweight wiring**\n\nWith Cortex, agents identify bottlenecks early, apply proven solutions, and adapt established workflows to new scenarios without context overload. Even as complexity increases, they maintain focus and improve with each use. Our deep research agent, powered by Cortex, **outperformed OpenAI's by 70% on enterprise search tasks and topped Microsoft's Deep Research benchmark.**\n\n## Who We're Looking For\n\n**We're partnering with enterprise teams building advanced agents that operate across complex business processes. Reach out** at [founders@aviro.ai](mailto:founders@aviro.ai) or book a time [here](https://cal.com/aviro/demo).\n\n![uploaded image](/media/?type=post&id=93860&key=user_uploads/2318974/b991d11e-fb8a-4323-a241-40bd0ad16860)\n\n", "url": "https://www.ycombinator.com/launches/OPs-aviro-continuous-upskilling-for-enterprise-ai-agents", "votes": 17, "created_at": "2025-09-18T14:56:08.191Z" }, { "id": 90937, "title": "Tropir: The AI that builds better AIs", "tagline": "You build the flow. We make it unbreakable.", "body": "## ➡️ Meet Tropir\n\nhttps://youtu.be/r9EajEr0tck\n\nModern AI pipelines fail constantly—hallucinations, bad retrievals, broken tool calls.\n\nTropir fixes that.\n\nIt traces exactly where things went wrong, suggests a targeted fix, reruns the pipeline, and tells you if the fix worked. All automatically.\n\n## 🔁 Trace → Fix → Rerun → Evaluate → Repeat\n\n![uploaded image](/media/?type=post&id=90937&key=user_uploads/2318975/1f24f662-f8c5-40f8-a5fa-6fbbc86ede7a)\n\nMost observability tools stop at logs. Tropir doesn’t just show you what broke—it tells you why, applies a fix, reruns the whole pipeline, and checks if the problem’s actually solved.\n\nYou can say: “Figure out why the @last_log wasn’t grounded in retrieved context. Use @faithfulness_eval.” Or just: “This output sucks. Use the full eval set to fix it.”\n\nTropir handles the rest.\n\n## 🧠 Why Tropir is Different\n\n* Traceback Engine — Finds the root cause across prompts, retrievals, tools, and context.\n* Fix + Rerun — Smart patches injected on the fly. No code rewrites.\n* Built-in Evals — Relevance, grounding, structure—all tested automatically. Add your own evals if you like!\n* Conversational Interface — It’s like interrogating your agent after runtime.\n* Plug & Play — Two lines of code to setup Tropir and you’re done.\n\n## 🔥 Try Tropir\n\n![uploaded image](/media/?type=post&id=90937&key=user_uploads/2318975/ca343c8f-0e54-427a-a05d-331bc3142703)\n\nTeams building RAG workflows, support bots, and research agents are shipping AI 60x faster with Tropir.\\\n→ [Book a demo](https://cal.com/aarush-kukreja/15min) to get early access!\\\n→ Visit [tropir.com](http://tropir.com) and sign up!", "url": "https://www.ycombinator.com/launches/Nej-tropir-the-ai-that-builds-better-ais", "votes": 21, "created_at": "2025-06-02T13:40:01.116Z" } ] }, { "id": 30247, "name": "Osmosis", "slug": "osmosis", "former_names": [ "Gulp" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/0e0e653e5da051b7c87a56070249ad76479b0594.png", "website": "https://osmosis.ai/", "all_locations": "San Francisco, CA, USA", "long_description": "Osmosis is a post-training platform that helps companies fine-tune language models using reinforcement learning. We work with fast-growing AI companies to train task/domain-specific models that beat foundation models on performance, cost, and latency. Our platform handles compute orchestration, reward modeling, and training run observability as a CLI-based product usable by developers and agents. ", "one_liner": "Reinforcement Learning (RL) for AI Agents", "team_size": 6, "industry": "B2B", "subindustry": "B2B", "launched_at": 1737656598, "tags": [ "Machine Learning", "Reinforcement Learning", "Infrastructure", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Winter 2025", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/osmosis", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2025/osmosis.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/osmosis-ai/", "x_url": "https://x.com/osmosis_ai", "fb_url": null, "cb_url": "https://www.crunchbase.com/organization/gulp-ai", "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/0e0e653e5da051b7c87a56070249ad76479b0594.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Brad Flora", "url": "https://www.ycombinator.com/people/brad-flora" }, "founders": [ { "user_id": 695591, "full_name": "Kasey Zhang", "title": "Founder/CEO", "founder_bio": "Currently: osmosis.ai \n\nPreviously: gaming founder, VC", "is_active": true, "linkedin_url": "https://linkedin.com/in/kasey-zhang", "x_url": "https://x.com/_WEEXIAO", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/89bfeaa626b0c7c6444e4634610ad1c228392c9f.jpg" }, { "user_id": 1479747, "full_name": "Andy Lyu", "title": "Founder/CTO", "founder_bio": "Defining how real time systems should be built. Previously worked on data architecture and ecosystem at TikTok recommendations as the 5th and youngest US hire", "is_active": true, "linkedin_url": "https://linkedin.com/in/baiqingl", "x_url": "https://x.com/viewerisland", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/4acaad1754ca9e5bf80ca3d25cdc86b8226565d6.jpg" } ], "jobs": [ { "id": 88480, "title": "Machine Learning Engineer ", "url": "https://www.ycombinator.com/companies/osmosis/jobs/N17l8oV-machine-learning-engineer", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Engineering", "role_type": "Machine learning", "salary_range": "$180K - $250K", "equity_range": "", "experience": "1+ years", "visa": "US citizen/visa only", "skills": [ "Reinforcement learning (RL)" ] } ], "news": [], "launches": [ { "id": 87475, "title": "Osmosis: Unlocking Real-Time Learning for AI Agents", "tagline": "Osmosis gives AI agents the ability to learn like DeepSeek R1, all in real-time ", "body": "What if AI agents could immediately learn from their mistakes?\n\n## TL;DR\n\nWhile AI agent capabilities continue to improve, the agentic workflows remain brittle since agents aren’t able to learn from tasks in real-time. Osmosis is a framework that helps your AI agents learn in real-time (similar to DeepSeek R1’s post-training loop) in order to make them better, faster, and more responsive.\n\nOn the first run, we scored 21.5% accuracy, which is 60% better than base 4o performance (16X more expensive!). **This means that even on a first run, we're 25X+ more cost effective than base 4o - and that would be significantly better on a second run.** We also averaged 7.5 steps per task while OpenAI Operator averaged 67 steps per task (faster and cheaper!).\n\nIf you’re building AI agents and want to make them better, faster, and more responsive, we’d love to chat! Email [kasey@gulp.ai](mailto:kasey@gulp.ai) or book time [here](https://calendly.com/kasey-gulp/).\n\n## The Problem\n\nAI agents still struggle with novel and complex (multi-step) tasks. Even OpenAI Operator (current state-of-the-art) just scores 58% accuracy on the WebArena benchmark. The worst part is that agents make the same mistakes again and again!\n\nAnd even when agents can solve a task, there’s generally a lack of optimization in their process. This ends up negatively affecting performance while also driving up inefficient inference costs for users.\n\n## The Solution\n\nOsmosis builds and maintains a ‘library’ of an agent’s previous interactions. It’s a super simple two line addition in a developer’s agent infrastructure that:\n\n* Ingests and understand a user’s prompt\n* Queries a vector database to search for similar past scenarios\n* Recommends a relevant plan of action for the task\n* Quantifies an agent’s performance when it finishes, and embeds the assessment\n\nJust like that, we have a flow that enables agents to learn in real-time with each successive task!\n\nHere’s a head-to-head (Operator versus Claude 3.5 Sonnet + Osmosis):\n\n\n\n**Osmosis was 3X faster and 6X more efficient in the number of steps taken!**\n\n## Our Ask\n\nIf you (or anyone you know) are building AI agents and struggling with accuracy, speed, or cost, we’d love to chat and see how we can help! Please reach out at [kasey@gulp.ai](mailto:kasey@gulp.ai), or book some time directly [here](https://calendly.com/kasey-gulp/).\n\n## The Team\n\nWe’ve (Kasey and Andy) known each other for over a decade!\n\n[Kasey](https://www.linkedin.com/in/kasey-zhang/) (CEO) previously co-founded and ran a gaming startup for two years before it was acquired. From there, he worked at various VC firms focusing on AI and data investments before deciding to jump back into building.\n\n[Andy](https://www.linkedin.com/in/baiqingl/) (CTO) previously worked at TikTok, where he joined as an early member of the data infrastructure team and eventually became a tech lead, running data infrastructure for recommendations across the core TikTok app, Shop, Live, and more.\n\n![uploaded image](/media/?type=post&id=87475&key=user_uploads/695591/d662049a-cf42-4d21-bc5f-08643fd3be5e)\n\n", "url": "https://www.ycombinator.com/launches/Mkt-osmosis-unlocking-real-time-learning-for-ai-agents", "votes": 155, "created_at": "2025-02-06T19:22:40.649Z" } ] }, { "id": 30254, "name": "HUD", "slug": "hud", "former_names": [ "Human Union Data", "hud", "HUD", "hud" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/53fffc4904f866d9d44ba9a47862b594f1d45667.png", "website": "https://www.hud.ai", "all_locations": "San Francisco, CA, USA", "long_description": "HUD is the platform for building high quality post training datasets. Over 50 businesses use HUD to build RL environments, sell them to AI labs, or train their own models from them.\r\n\r\nOur mission is to enable a generation of data entrepreneurs. The previous generation built apps to impact the world. We believe people will build infrastructure around data, both digital and physical, that align AIs to their specific goals.", "one_liner": "Platform and marketplace for RL environments and AI data", "team_size": 24, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1740930546, "tags": [ "Marketplace", "Reinforcement Learning", "B2B", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Winter 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada", "Remote", "Partly Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/hud", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2025/hud.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/hud-evals", "x_url": "https://x.com/hud_evals", "fb_url": null, "cb_url": null, "github_url": "https://github.com/hud-evals", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/53fffc4904f866d9d44ba9a47862b594f1d45667.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Aaron Epstein", "url": "https://www.ycombinator.com/people/aaron-epstein" }, "founders": [ { "user_id": 738085, "full_name": "Jay Ram", "title": "Founder/CEO", "founder_bio": "CEO @ HUD\n\nprev: consumer apps, ml research, quant research", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/jay-ram-29003b198/", "x_url": "https://x.com/jayendra_ram", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/fa944fcd27a700e758b64d952d5d6b56b8c09959.jpg" }, { "user_id": 1308385, "full_name": "Lorenss Martinsons", "title": "Founder", "founder_bio": "Co-CEO @ hud\ninterested in natural intelligence, flowers, and the human condition", "is_active": true, "linkedin_url": "https://linkedin.com/in/lorenss", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/5c1969855011e9ec552ade11253fc6a178daa3f4.jpg" }, { "user_id": 353935, "full_name": "Parth Patel", "title": "Founder/CTO", "founder_bio": "CTO @ HUD.\n[a-z]*", "is_active": true, "linkedin_url": "https://linkedin.com/in/parth220", "x_url": "https://twitter.com/parth220", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/b14c856200f7739c08dc25aa5ea56754092925ce.jpg" } ], "jobs": [ { "id": 93541, "title": "Research Engineer (General)", "url": "https://www.ycombinator.com/companies/hud/jobs/9sXeoEj-research-engineer-general", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 93543, "title": "GTM Engineer", "url": "https://www.ycombinator.com/companies/hud/jobs/S4iMOoe-gtm-engineer", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $180K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 93544, "title": "Talent Engineer", "url": "https://www.ycombinator.com/companies/hud/jobs/gkaIWwX-talent-engineer", "location": "San Francisco", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $165K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 93545, "title": "Marketing & Events Lead ", "url": "https://www.ycombinator.com/companies/hud/jobs/B59vJPJ-marketing-events-lead", "location": "San Francisco", "type": "Full-time", "role": "Marketing", "role_type": null, "salary_range": "$135K - $205K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 98124, "title": "Growth Lead", "url": "https://www.ycombinator.com/companies/hud/jobs/HeJ5jdh-growth-lead", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$165K - $235K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 99313, "title": "Research Engineer, Synthetic Data", "url": "https://www.ycombinator.com/companies/hud/jobs/FZ59eeq-research-engineer-synthetic-data", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 99315, "title": "Forward Deployed Research Engineer", "url": "https://www.ycombinator.com/companies/hud/jobs/Ve0lhgX-forward-deployed-research-engineer", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $180K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 99316, "title": "Research Engineer, Benchmarks", "url": "https://www.ycombinator.com/companies/hud/jobs/TXcQo9I-research-engineer-benchmarks", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 99317, "title": "Research Engineer, QC Automation", "url": "https://www.ycombinator.com/companies/hud/jobs/iRcj3ZI-research-engineer-qc-automation", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 101991, "title": "Lead Research Engineer, Data Quality", "url": "https://www.ycombinator.com/companies/hud/jobs/95U7bba-lead-research-engineer-data-quality", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$120K - $295K", "equity_range": "", "experience": "5+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 105817, "title": "Senior Software Engineer, Reinforcement Learning", "url": "https://www.ycombinator.com/companies/hud/jobs/xRmqcxW-senior-software-engineer-reinforcement-learning", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $225K", "equity_range": "", "experience": "3+ years", "visa": "Will sponsor", "skills": [] }, { "id": 108980, "title": "Senior Security Engineer", "url": "https://www.ycombinator.com/companies/hud/jobs/fi9ZRlo-senior-security-engineer", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$105K - $260K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 110432, "title": "Open Role", "url": "https://www.ycombinator.com/companies/hud/jobs/06bdplz-open-role", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$100K - $250K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 110433, "title": "Senior Software Engineer, Infrastructure", "url": "https://www.ycombinator.com/companies/hud/jobs/Ul0tPP8-senior-software-engineer-infrastructure", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$105K - $260K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 110434, "title": "Delivery Engineer", "url": "https://www.ycombinator.com/companies/hud/jobs/i5FQmnq-delivery-engineer", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$100K - $225K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 110435, "title": "Research Engineer, Privacy and Anonymization", "url": "https://www.ycombinator.com/companies/hud/jobs/dFaIN6u-research-engineer-privacy-and-anonymization", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 111723, "title": "Business Development Lead", "url": "https://www.ycombinator.com/companies/hud/jobs/xNf2MQx-business-development-lead", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 111724, "title": "GTM Lead, Commercial Partnerships", "url": "https://www.ycombinator.com/companies/hud/jobs/S99Rxu9-gtm-lead-commercial-partnerships", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 111725, "title": "Head of Talent", "url": "https://www.ycombinator.com/companies/hud/jobs/WvW8msz-head-of-talent", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$140K - $210K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 111726, "title": "Business Operations", "url": "https://www.ycombinator.com/companies/hud/jobs/fsfQq2Z-business-operations", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$100K - $165K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 112120, "title": "Product Manager, Marketplace", "url": "https://www.ycombinator.com/companies/hud/jobs/iKaUFbG-product-manager-marketplace", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Product", "role_type": null, "salary_range": "$100K - $225K", "equity_range": "", "experience": "5+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 112121, "title": "Software Engineer, UI/UX", "url": "https://www.ycombinator.com/companies/hud/jobs/ePuzeUI-software-engineer-ui-ux", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $225K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 112122, "title": "Product Manager, Platform", "url": "https://www.ycombinator.com/companies/hud/jobs/Jnj6nf6-product-manager-platform", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Product", "role_type": null, "salary_range": "$100K - $225K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 112123, "title": "Senior Research Manager", "url": "https://www.ycombinator.com/companies/hud/jobs/jeosyh4-senior-research-manager", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$120K - $295K", "equity_range": "", "experience": "5+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 112792, "title": "Research Engineer, Robotics Data", "url": "https://www.ycombinator.com/companies/hud/jobs/nC83OjE-research-engineer-robotics-data", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 113103, "title": "Product Engineer", "url": "https://www.ycombinator.com/companies/hud/jobs/j7SbPKn-product-engineer", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $225K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 113411, "title": "Customer Operations Lead", "url": "https://www.ycombinator.com/companies/hud/jobs/bUCWq7i-customer-operations-lead", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$100K - $165K", "equity_range": "", "experience": "5+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 113798, "title": "Research Engineer, Robotics Evals", "url": "https://www.ycombinator.com/companies/hud/jobs/SToEZ1D-research-engineer-robotics-evals", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 113799, "title": "Research Engineer, Robotics Evals", "url": "https://www.ycombinator.com/companies/hud/jobs/CCvHy70-research-engineer-robotics-evals", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $230K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 116141, "title": "Technical Writer", "url": "https://www.ycombinator.com/companies/hud/jobs/jllqbXF-technical-writer", "location": "San Francisco / Singapore / Remote (Asia) / Remote (North America)", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$100K - $165K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] } ], "news": [], "launches": [] }, { "id": 30278, "name": "Agentin AI", "slug": "agentin-ai", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/da9bdd65896a691965e96e56efdfaa0880fcac0e.png", "website": "https://www.agentin.ai", "all_locations": "San Francisco, CA, USA", "long_description": "", "one_liner": "Foundational model research", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1740032946, "tags": [ "Reinforcement Learning", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2025", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/agentin-ai", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2025/agentin-ai.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "http://linkedin.com/company/agentin-ai", "x_url": "https://x.com/ai_agentin", "fb_url": null, "cb_url": "https://www.crunchbase.com/organization/de1aaf20-8ab6-4c76-bf26-fb6dfbb2d0de", "github_url": "https://github.com/agentin-ai", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/da9bdd65896a691965e96e56efdfaa0880fcac0e.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/67234eea6d230948adda5840a5b4f2210d2fd67e.png", "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/f93fcd1962726b7c4845afc71c02d000d90d3399.png" ], "primary_partner": { "name": "Harj Taggar", "url": "https://www.ycombinator.com/people/harj-taggar" }, "founders": [ { "user_id": 423828, "full_name": "Sankeerth Rao Karingula", "title": "Founder/CEO", "founder_bio": "Sankeerth is the Founder and CEO of Agentin AI, an AI research lab training foundation models to conduct original research.\n\nPreviously a Research Scientist at Google, he holds a PhD in Machine Learning from UC San Diego and studied Electrical Engineering at IIT Bombay.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/sankeerthrao", "x_url": "https://x.com/sankeerth1729", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/581460eed0156434e9f8b1730921d9e331551c71.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 88196, "title": "Agentin AI - Training Foundation Models for Research", "tagline": "An AI research lab training models to conduct original research.", "body": "We are building foundation models designed to work on problems without known answers.\n\nOur goal is to train models that can explore ideas, form hypotheses, test approaches, and build on their findings to make new discoveries.", "url": "https://www.ycombinator.com/launches/MwW-agentin-ai-training-foundation-models-for-research", "votes": 11, "created_at": "2025-03-03T05:53:49.367Z" } ] }, { "id": 30289, "name": "Sherpa Labs", "slug": "sherpa-labs", "former_names": [ "DagHub.ai" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/9fc6e69fe5c9b9ed32dbb15b81f5057c0f2e4abd.png", "website": "https://www.sherpalabs.ai", "all_locations": "San Francisco, CA, USA", "long_description": "", "one_liner": "Frontier Data", "team_size": 5, "industry": "B2B", "subindustry": "B2B", "launched_at": 1737526775, "tags": [ "Reinforcement Learning", "Data Labeling", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2025", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/sherpa-labs", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2025/sherpa-labs.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/sherpalabs-ai", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/9fc6e69fe5c9b9ed32dbb15b81f5057c0f2e4abd.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Tom Blomfield", "url": "https://www.ycombinator.com/people/tom-blomfield" }, "founders": [ { "user_id": 2275056, "full_name": "Justin Jiang", "title": "Founder", "founder_bio": "Building frontier RL data pipelines @ Sherpa Labs. Ex-research scientist from Surge, ex-quant researcher from Millennium, Vatic, Optiver, working on medium frequency statistical arbitrage and high frequency trading algorithms. Data scientist at FedEx and Tradeweb. Graduated from Harvey Mudd with a double BS in Math and CS.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/justinjiang12/", "x_url": "https://x.com/jostinjiang", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e4bfe9f614fc52837827a9aa23efb42eddd4c4af.jpg" } ], "jobs": [], "news": [], "launches": [] }, { "id": 30350, "name": "TrainLoop", "slug": "trainloop", "former_names": [ "Phoenix" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/4fad47355d59021d6d855bc4b2dc95cbc0ef2f1a.png", "website": "http://trainloop.ai", "all_locations": "San Francisco, CA, USA", "long_description": "TrainLoop makes it effortless for developers to supercharge LLM performance through reinforcement learning.", "one_liner": "Reasoning Fine-Tuning", "team_size": 6, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1738700190, "tags": [ "Developer Tools", "Generative AI", "Reinforcement Learning" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Winter 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/trainloop", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2025/trainloop.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/trainloopai/", "x_url": "https://x.com/TrainLoop_ai", "fb_url": null, "cb_url": null, "github_url": "https://github.com/TrainLoop", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/4fad47355d59021d6d855bc4b2dc95cbc0ef2f1a.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "David Lieb", "url": "https://www.ycombinator.com/people/david-lieb" }, "founders": [ { "user_id": 1997228, "full_name": "Jackson Stokes", "title": "Founder/CEO", "founder_bio": "Cofounder at TrainLoop, previously Google Gemini & AlphaFold performance ", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/jacksonstokes/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/6830e129e7460fce6375e8137b2a667fbe01fa75.jpg" }, { "user_id": 681927, "full_name": "Mason Pierce", "title": "Founder/CTO", "founder_bio": "My background is in applying AI models to business use-cases and have always found fine-tuning to be way harder than it should be - My mission is to unlock fine-tuning for all builders :)", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/malapier", "x_url": "https://x.com/mlpierce22", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e0708b326a808512f74417ba49da6268c9d5777f.jpg" } ], "jobs": [ { "id": 116537, "title": "Head of Enterprise", "url": "https://www.ycombinator.com/companies/trainloop/jobs/elVnWkD-head-of-enterprise", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$150K - $250K", "equity_range": "0.50% - 1.50%", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 116538, "title": "Research Engineer", "url": "https://www.ycombinator.com/companies/trainloop/jobs/WABE7WW-research-engineer", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Engineering", "role_type": "Machine learning", "salary_range": "$150K - $250K", "equity_range": "0.50% - 1.50%", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 116539, "title": "Research Scientist", "url": "https://www.ycombinator.com/companies/trainloop/jobs/1ni2P0Y-research-scientist", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Science", "role_type": "Research", "salary_range": "$150K - $250K", "equity_range": "0.50% - 1.50%", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 116540, "title": "UX / Product Designer", "url": "https://www.ycombinator.com/companies/trainloop/jobs/OtdpMKl-ux-product-designer", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Design", "role_type": "Product design", "salary_range": "$150K - $250K", "equity_range": "0.50% - 1.50%", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [] } ], "news": [], "launches": [ { "id": 87957, "title": "TrainLoop: Unlock Next-Level Reasoning through Fine-Tuning", "tagline": "Eliminate unwanted responses and match ideal outputs for your product. ", "body": "# Unreliable RAG or code generation? We can help.\n\nReasoning models have been all the rage lately because they beat generic benchmarks. The problem is that your business isn’t a generic benchmark - it’s a set of specific vertical tasks like codegen, compliance, legal or healthcare. Massive companies like Google and OpenAI have internal tools to train their models, but those aren’t available to the people that need it: the developers deploying these models into production.\n\nWe’ve personally been involved on both sides: Jackson optimized the Gemini models at Google and Mason hit the limits of off shelf models while leading engineering at Second (YC W23).\n\nSo we created **TrainLoop**, packaging the same RL techniques big AI labs use into an accessible platform. Our process is three simple steps:\n\n1. **Data Curation:** Our lightweight SDK (just three lines of code) gathers training signals from actual usage.\n2. **Training:** We build a reward model that teaches your LLM what output you prefer.\n3. **Inference:** Deploy automatically and call your model via standard APIs.\n\n\n\n# Ready to Level Up Your Model?\n\nIt’s time to move past “prompt-hell” and unreliable outputs. [**Join our alpha**](https://app.trainloop.ai/waitlist) to make your language model an expert in your business and unlock production-ready performance.", "url": "https://www.ycombinator.com/launches/Msf-trainloop-unlock-next-level-reasoning-through-fine-tuning", "votes": 32, "created_at": "2025-02-24T16:02:19.658Z" } ] }, { "id": 30352, "name": "Monte", "slug": "monte", "former_names": [ "Fluidize" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/213b59f051867cd51ae0008db2962b11e76fd9f3.png", "website": "https://trymonte.ai", "all_locations": "San Francisco, CA, USA", "long_description": "Monte is an applied research lab building products for continual learning and recursive self-improvement for AI agents.", "one_liner": "Continual Learning for Agents", "team_size": 3, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1753994250, "tags": [ "Reinforcement Learning", "B2B", "Infrastructure", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/monte", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2025/monte.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/monte-ai", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/213b59f051867cd51ae0008db2962b11e76fd9f3.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jared Friedman", "url": "https://www.ycombinator.com/people/jared-friedman" }, "founders": [ { "user_id": 1442725, "full_name": "Henry Bae", "title": "Founder/CEO", "founder_bio": "Co-Founder @ Monte. He has an AB/SB/SM in Physics, CS, and Applied Math at Harvard.", "is_active": true, "linkedin_url": "https://linkedin.com/in/henrybae", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/3a08df12c5c2e00192e906f21e47b8490974d058.jpg" }, { "user_id": 1874285, "full_name": "Alex Fleury", "title": "Founder/COO", "founder_bio": "Co-founder @ Monte | CS, statistics, AI research @ Harvard | 4 minute miler", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/victoralexfleury/", "x_url": "https://x.com/alexfleury160", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/b8135fb60a6325f6bf506ca77edef9aaab45f366.jpg" }, { "user_id": 1599108, "full_name": "Jamin Liu", "title": "Founder", "founder_bio": "Co-founder @ Monte | CS & MCB @ Harvard", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/jamin-liu-255277163", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/6a757051e43e5cee1744f246cd8ad8ca651d2b13.jpg" } ], "jobs": [], "news": [], "launches": [] }, { "id": 30462, "name": "Freesolo", "slug": "freesolo", "former_names": [ "Linkd", "Clado", "Freesolo", "Clado" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/45d118ce06f810c00f7b1f3105cf4f7906bd0d55.png", "website": "https://freesolo.co", "all_locations": "San Francisco, CA, USA", "long_description": "Freesolo works with companies to encode user-trajectory knowledge into specialized models that outperform the state of the art at much lower latency and cost.", "one_liner": "Product Native Models", "team_size": 4, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1742722403, "tags": [ "Reinforcement Learning", "B2B", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Spring 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/freesolo", "api": "https://devasheeshg.github.io/yc-api/batches/spring-2025/freesolo.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/freesoloco", "x_url": "https://x.com/freesoloco", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/45d118ce06f810c00f7b1f3105cf4f7906bd0d55.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Pete Koomen", "url": "https://www.ycombinator.com/people/pete-koomen" }, "founders": [ { "user_id": 1528922, "full_name": "Eric Mao", "title": "Founder/CEO", "founder_bio": "Hello, my name is Eric. I used to study Computer Science and Business at the University of Pennsylvania. \n", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/eric-mao/", "x_url": "https://x.com/ericmao", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/91bbe42a29bedc50c83a6cf9304171641142c6f1.jpg" }, { "user_id": 2374036, "full_name": "Tom Zheng", "title": "Founder/CTO", "founder_bio": "Cofounder @ Freesolo, automating model training. ", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/tomzhengy", "x_url": "https://x.com/tomzheng", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/d09ed18ad61b20f68761ea271765e500c74d7262.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 105027, "title": "Freesolo: Post-Training Built for your Agent", "tagline": "Small models, frontier performance", "body": "Not every AI interaction is best served with a large frontier model. There is a long tail of trillion-token use cases, from tagging to search, best served by a sub-10B-parameter model that runs in milliseconds and costs many orders of magnitude less than the frontier. However, engineers historically had to choose between model size and quality; as models got smaller, performance, adherence, and recall dropped linearly. Post-training on production data closes that gap. We built Freesolo Flash to make training loops like SFT and RL easy and end-to-end completable through your coding agent.\n\nWe accomplish this in a few different ways\n\n* Upfront pricing: instead of billing by GPU hours or tokens spent while training, we quote the cost of the entire run upfront, so your agent can accurately tweak the dataset, model size, and algorithms it uses while staying in your budget before it starts the run.\n* Our GPU infrastructure is optimized to make your specific run as in-expensive and fast as possible. This optimization means training with flash is 8x less expensive for SFT and 5.5x less expensive for GRPO (RL) when compared to Tinker.\n\nFlash is built out of our own frustrations with current managed post-training solutions, especially for SLMs. We believe that unlocking frontier capability for a narrow task into a small model will prove to be the best improvement for all agentic product UX. Flash is our first step towards solving this. Just grab a Freesolo API Key, point your agent at the training package, and watch it push your lightweight model beyond the frontier.\n\nFlash is publicly available today at  [ ](https://freesolo.co/). Train on!", "url": "https://www.ycombinator.com/launches/RJz-freesolo-post-training-built-for-your-agent", "votes": 3, "created_at": "2026-07-13T19:38:20.322Z" }, { "id": 90716, "title": "Clado: Deep Research for People", "tagline": "100,000+ agents to find the right people at the right time", "body": "Clado puts 200 million+ people profiles (and growing) at your fingertips with simple, agentic search. \n\nIf you’re interested in a global people search for sales, spreadsheets, hiring, and more, let’s talk: https://cal.com/clado/chat\n\n**We’d love for you all to try Clado out live!** https://clado.ai\n\nhttps://youtu.be/NEqtIFNZJoU\n\n❌ The Problem: Keywords, Filters, Data\n\nFinding the right people is broken. Current methods of people search are made up of endless filter menus, incomplete data sets, and keyword search. It takes tremendous effort and time to find people for sales and hiring, as current search methods rely on fuzzy filters in limited datasets. \n\n✅ Solution: Clado uses agentic search to find exactly who you're looking for. Just describe the qualities you want, and we will deploy hundreds of thousands of agents to find them.\n\n![uploaded image](/media/?type=post&id=90716&key=user_uploads/2374036/2faf2b26-e574-43af-8bdc-933ad4b2f2e7)\n\nThese example searches are time-consuming to execute with standard filtering tools and elude traditional keyword filters.\n\n_\"AI Engineers I can poach for my pharmaceutical startup/interested in biology in Copenhagen.\"_\n\nAn even harder one:\n\n_“Every exec (Owner/CEO/COO/CFO/Partner/Head of Growth/CTO) at accounting firms in the US and UK between 250 - 2000 employees\"_\n\nAfter internally benchmarking our people search engine against other platforms, our model achieves SOTA performance. \\\n\\\nWhat good is finding the right person if you can’t easily contact them? We’re also able to achieve SOTA performance on contact delivery information retrieval given any Linkedin URL (provided for every result in our search). \n\n![](https://lh7-rt.googleusercontent.com/docsz/AD_4nXdlMEsme5O2KS7jDjVQ5meLUOG0wSsn4dUDYSaIlwQ2hYDg2Sa7HGCVi_7rjnEAfXWNdODe05Fhghlf6BVYD0z6QlbEjh4B1N1rGLlh3M_RqfIRDrIdpkdKJQzZivA3A7cMH7D93w?key=IzTsz7ArVY5QuH96a3WVtw)\n\nTesting Conditions:[ https://pastebin.com/98PVVN9b](https://pastebin.com/98PVVN9b)\\\n\\\nThe Team:\\\nTom (UCSD) & Eric (UPenn M&T) are both ambitious freshmen who met in elementary school. They previously founded a nonprofit and won Canada’s largest data science & physics competitions.\\\n\\\nThey’ve also sailed over 300 nautical miles over 10 days, canoed 100km over 6 days, walked around Toronto hanging up posters for their first business for two weeks straight, and traveled to Sri Lanka on a three-day notice to present at a conference together. \\\n\\\nTry clado: https://clado.ai\n\n📅 Book a meeting: [**https://cal.com/clado/chat**](https://cal.com/clado/chat)\n\n📧 Email us: [**founders@clado.ai**](mailto:founders@clado.ai)", "url": "https://www.ycombinator.com/launches/NbA-clado-deep-research-for-people", "votes": 80, "created_at": "2025-05-27T20:44:55.352Z" } ] }, { "id": 30488, "name": "Theta", "slug": "theta", "former_names": [ "Handler AI", "Theta Software" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/bef5c2a34194462b75e30c574accec7a94eef920.png", "website": "https://thetasoftware.com", "all_locations": "San Francisco, CA, USA", "long_description": "", "one_liner": "Specialized AI for Every Job", "team_size": 7, "industry": "B2B", "subindustry": "B2B", "launched_at": 1746731099, "tags": [ "Reinforcement Learning", "B2B", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Spring 2025", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/theta", "api": "https://devasheeshg.github.io/yc-api/batches/spring-2025/theta.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/try-theta-software/", "x_url": "https://x.com/trytheta", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/bef5c2a34194462b75e30c574accec7a94eef920.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Garry Tan", "url": "https://www.ycombinator.com/people/garry-tan" }, "founders": [ { "user_id": 537913, "full_name": "Rayan Garg", "title": "Founder/CEO", "founder_bio": "Cofounder @ Theta", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/rayan-garg/", "x_url": "https://x.com/RayanGarg", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/c149d47b48ed70e9578a05089862fe95b2a34f09.jpg" }, { "user_id": 1027108, "full_name": "Tanmay Sharma", "title": "Founder/CPO", "founder_bio": "Co-founder @ Theta", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/tanmayxsharma/", "x_url": "https://x.com/tsha444", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/587a5a95d8b018cf50a39e232bd35ba6813223fc.jpg" }, { "user_id": 1962783, "full_name": "Gurvir Singh", "title": "Founder/CTO", "founder_bio": "Co-founder @ Theta", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/gurvir-s/", "x_url": "https://x.com/_gurvir_", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/ba15e188b5b6aa70b51f319fdb6e33bb73acb28c.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 90230, "title": "Theta: Self-Learning for AI Agents", "tagline": "An Intelligent Memory Layer for Real-time Learning", "body": "### **TL;DR**\n\nTheta is building self-learning and real-time adaptation for AI agents.\n\nWe’re starting with an intelligent memory layer so agents can remember and learn from previous interactions. With a simple four-line addition to your existing code, our memory layer uses real-time learning to analyze every run for mistakes and optimizations. The relevant insights are then passed on to your agent during future runs. \n\nWe’ve already improved the accuracy of OpenAI Operator by 43% with 7x fewer steps taken. If you’re building AI agents where reliability and speed are priorities, book a meeting [here](https://calendly.com/d/cndj-84h-9qz) or contact us at [founders@thetasoftware.ai](mailto:founders@thetasoftware.ai).\n\n\\\nhttps://youtu.be/TRNWfhyUI4E\n\n### **The Current Approach**\n\nAgents struggle to adapt to complex, real-world workflows. Workflows are dynamic, but agents remain static.\n\n* They get stuck in loops, repeating mistakes and requiring constant human guidance in the right direction. Your prompts get longer and longer as you try to hard-code solutions.\n* Even when your agent successfully completes a task, this long process starts all over again for any future run. Starting a new Cursor chat or Operator session means losing all of the insights and “aha moments” your agent had.\n\nLearning is fundamentally iterative, but agents can’t learn because they have no memory across runs.\n\n### **How Theta Works**\n\nTheta is building the infrastructure for agents to self-learn and adapt in real-time. The first component is an intelligent memory layer that learns from your agent’s previous runs. Just add four lines of code to your agent stack to get started:\n\n* At the end of the current run, Theta generates and embeds an analysis of your agent’s trajectory. It identifies critical steps and assesses overall performance.\n* Before the next run, Theta generates a detailed plan based on the specified task. It gathers relevant insights and optimizations based on previous runs.\n* With real-time learning, the memory layer continues to improve with more runs—meaning your agent gets better and better with zero human intervention.\n\nUsing this memory layer, we were able to improve the accuracy of OpenAI Operator by 43%. With optimized trajectories, Operator also took 7x fewer steps, resulting in better speed and cost.\n\n### **Our Ask**\n\nIf you’re building agents that need to perform dynamic, real-world workflows at the highest accuracy and speed, reach out to [founders@thetasoftware.ai](mailto:founders@thetasoftware.ai) or book some time [here.](https://calendly.com/d/cndj-84h-9qz)\n\n### **Our Team**\n\nRayan has previously done ML research as Head of Product at DeepSilicon. He’s been childhood friends with Tanmay since third grade, who previously built an AI browser and developed browser agents at MultiOn. During Rayan’s freshman year of college, he met Gurvir, who built distributed ML systems at Cornell, focusing on post-training and RL.", "url": "https://www.ycombinator.com/launches/NTK-theta-self-learning-for-ai-agents", "votes": 48, "created_at": "2025-05-08T22:13:52.982Z" } ] }, { "id": 30501, "name": "Cartpole", "slug": "cartpole", "former_names": [ "Prophet", "Jazzberry" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/154693ac9eca92a2d8f93ead7153a0858c38d2b3.png", "website": "https://cartpole.com", "all_locations": "San Francisco, CA, USA", "long_description": "We're creating reinforcement learning environments for training frontier models.", "one_liner": "Building reinforcement learning environments", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1744600545, "tags": [ "Reinforcement Learning", "Data Labeling", "AI", "ML" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Spring 2025", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/cartpole", "api": "https://devasheeshg.github.io/yc-api/batches/spring-2025/cartpole.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/cartpole", "x_url": "https://x.com/cartpoleai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/154693ac9eca92a2d8f93ead7153a0858c38d2b3.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Diana Hu", "url": "https://www.ycombinator.com/people/diana-hu" }, "founders": [ { "user_id": 2557020, "full_name": "Mateo Perez", "title": "Founder/CEO", "founder_bio": "Building Cartpole. PhD @ CU Boulder.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/mateo-perez/", "x_url": "https://x.com/m_perez_", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/bc4e196cc3b6bc16eb7775444008cd50ad1ddd64.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 90047, "title": "Jazzberry - AI bug finding with real code execution", "tagline": "We're creating the best AI agent for bug finding", "body": "Hello everyone, \n\nWe’re Mateo and Marco, and we're excited to introduce Jazzberry, an AI agent for finding bugs. Jazzberry is powered by running your code in a secure sandbox and integrates directly into your pull request workflow on GitHub.\n\n**Try it here →** \n\n\n\n**Why we built Jazzberry**\n\nAI code generation has dramatically increased the amount of code that is written without the tools for bug finding keeping up, leading to a world with more buggy code that is hindering developers. To improve bug finding, we’ve created an AI agent with the ability to run snippets of your code. We're not just flagging potential issues based on code patterns; we're observing the real behavior of your software. Here’s how it works.\n\n1. When you create a pull request, Jazzberry gets to work.\n2. Your repository is cloned into a microVM.\n3. Our agent examines the contents of the pull request to guide its search. It can write and execute code in the microVM as it looks for bugs.\n4. Jazzberry returns a clear markdown table directly as a comment.\n\n![uploaded image](/media/?type=post&id=90047&key=user_uploads/2557020/31d75b00-abc3-4a60-8ef5-b3a685edfcd5)\n\n**Real Bugs, Real Impact**\n\nWe've already seen Jazzberry make a tangible difference for our early users. So far, Jazzberry has…\n\n* successfully identified a SQL injection vulnerability\n* discovered a critical logical error in API key validation\n* [found an incorrect page navigation](https://blog.openint.dev/how-we-scale-prs/) (see below, we are formerly known as Prophet)\n\n![uploaded image](/media/?type=post&id=90047&key=user_uploads/2557020/864328cf-c4b9-4b9a-b267-e9059a158f32)\n\n**Who we are**\n\nWe have been friends for over a decade, and are coming together to create Jazzberry. Marco completed his MS at the University of Stellenbosch specializing in LLMs for automated test suite generation. Mateo completed his PhD at the University of Colorado Boulder with a research focus on reinforcement learning and formal methods.\n\n**Our Ask**\n\n* **Try it out** – Jazzberry is available now with a free trial at .\n* **Contact us** – We’d love to hear your feedback and what bugs Jazzberry helps you catch. Reach us at [founders@jazzberry.ai](mailto:founders@jazzberry.ai).", "url": "https://www.ycombinator.com/launches/NQN-jazzberry-ai-bug-finding-with-real-code-execution", "votes": 42, "created_at": "2025-05-05T22:37:45.137Z" } ] }, { "id": 30521, "name": "Kairos", "slug": "kairos", "former_names": [ "Janus" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/3b44f7b050ecc3ee1451baeb4e0a733b19937955.png", "website": "https://www.withkairos.co/", "all_locations": "San Francisco, CA, USA", "long_description": "Kairos closes the last-mile reliability gap in AI deployments. \r\n\r\nWe bring frontier techniques to companies in critical industries, deploying specialized agents that encode operator expertise and reliably automate their most manual workflows.", "one_liner": "Specialized AI for Critical Industries", "team_size": 3, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1748466933, "tags": [ "AIOps", "Machine Learning", "Reinforcement Learning", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Spring 2025", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/kairos", "api": "https://devasheeshg.github.io/yc-api/batches/spring-2025/kairos.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/withkairos/", "x_url": "https://x.com/with_kairos", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/3b44f7b050ecc3ee1451baeb4e0a733b19937955.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Andrew Miklas", "url": "https://www.ycombinator.com/people/andrew-miklas" }, "founders": [ { "user_id": 1090038, "full_name": "Shivum Pandove", "title": "Founder", "founder_bio": "Kairos | Prev. ML/CS at Carnegie Mellon, Scaled 3+ startups as a SWE/PM from 0 -> 1, and conducted DL research across comp. bio labs", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/shivump1/", "x_url": "https://x.com/shivump101", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/8567c0d26b02bc6e0b04899282d36f727db954a9.jpg" } ], "jobs": [], "news": [], "launches": [] }, { "id": 30629, "name": "Topological", "slug": "topological", "former_names": [ "TipTop", "TipTop AI" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/2b9c845239fb1f16317889a4a843409e109b24f6.png", "website": "https://www.topological.com", "all_locations": "San Francisco, CA, USA", "long_description": "Topological is developing physics-based foundation models for CAD optimization. We help hardware teams iterate at the same speed that software teams do. Our technology is accelerating the engineering workflow with AI and scales design and optimization to identify the ideal designs for complex problems given their physical constraints with enhanced speed and performance. \r\n\r\nOur first model, UToP-v1, is a SOTA topology optimization model that understands physics, geometry, and manufacturability. It can generate the most efficient design given a problem’s physical requirements. It has <5% compliance error and is 1930x faster than current methods. We're reimagining mechanical engineering and computational design with precision spatial AI.", "one_liner": "Physics-based foundation models for CAD optimization.", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1755729176, "tags": [ "Reinforcement Learning", "Robotics", "Design Tools", "3D Printing", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/topological", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2025/topological.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/topologicalai", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/2b9c845239fb1f16317889a4a843409e109b24f6.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Brad Flora", "url": "https://www.ycombinator.com/people/brad-flora" }, "founders": [ { "user_id": 2395628, "full_name": "Nicole Ma", "title": "Founder/CEO", "founder_bio": "Building Topological (YC S25). Prev. CS/MechE @ Stanford, physics simulations @ NASA", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ma06nicole/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/a27a3d8667b1bb56b79f33bd09bed9589db509ef.jpg" }, { "user_id": 2391326, "full_name": "Zeyneb Kaya", "title": "Founder/CTO", "founder_bio": "Building Topological (YC S25). Prev CS/Physics/Math @ Stanford; research in RL, pretraining, efficient algorithms, synthetic data, physics. ", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/zeyneb-kaya/", "x_url": "https://x.com/zeynebnkaya", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/1dd98978718ff1e9352d4f96ba5623a1a68ad954.jpg" } ], "jobs": [], "news": [], "launches": [] }, { "id": 30643, "name": "Nuntius", "slug": "nuntius", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/419ca938d39201c4fff4ef02cd003dd413be8954.png", "website": "https://www.nuntius.ai", "all_locations": "San Francisco, CA, USA", "long_description": "Applying artificial intelligence to economically productive use-cases requires bulletproof alignment. We’re solving that.", "one_liner": "Making Models Follow Rules.", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1754692935, "tags": [ "Reinforcement Learning", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2025", "status": "Inactive", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/nuntius", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2025/nuntius.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/106397474", "x_url": "https://x.com/nuntiusai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/419ca938d39201c4fff4ef02cd003dd413be8954.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/eb253a217d35079a3c96f27d3240a70e45577b4e.png", "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/e1e2ec16f39cccdfb05abdfd0e05929ec1624e05.png" ], "primary_partner": { "name": "Jon Xu", "url": "https://www.ycombinator.com/people/jon-xu" }, "founders": [], "jobs": [], "news": [], "launches": [] }, { "id": 30730, "name": "idler", "slug": "idler", "former_names": [ "Idler" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/3ab3b751ff1a6569beaba5e0f91c3c4dbcda3e55.png", "website": "https://idler.ai/", "all_locations": "San Francisco, CA, USA", "long_description": "idler is a frontier data research lab. We build the evals and environments that the world's leading frontier labs use to measure and train their models.", "one_liner": "Hill-climb any benchmark, for any model, in any domain.", "team_size": 12, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1749750449, "tags": [ "Reinforcement Learning" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Summer 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/idler", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2025/idler.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/idler/", "x_url": "https://x.com/idler_ai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/3ab3b751ff1a6569beaba5e0f91c3c4dbcda3e55.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/05f47d6348a4ea7e749e66a27b4ffc959061b24e.png" ], "primary_partner": { "name": "Harj Taggar", "url": "https://www.ycombinator.com/people/harj-taggar" }, "founders": [ { "user_id": 316251, "full_name": "Ivan Chub", "title": "Founder/CEO", "founder_bio": "idler. Previously cofounder of Dark Forest, 0xPARC, Zupass, Hack Lodge; also previously at Facebook, Dynasty, AppSheet.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ivanchub/", "x_url": "https://x.com/chubivan", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e3627b353f012ad57dd7503754947dca63c3b803.jpg" }, { "user_id": 557067, "full_name": "Nalu Concepcion", "title": "Founder", "founder_bio": "currently: scaling RL environments. \n\nformerly: orchestrated 500+ person parties, monthly @ notsominimoments · swe @ msft · product @ poshmark · data science @ paymongo", "is_active": true, "linkedin_url": "https://linkedin.com/in/nalu-concepcion/", "x_url": "https://x.com/naluconcepcion", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/2c1b57f71dcf64e903abfa38471c9fe3126aabbb.jpg" }, { "user_id": 1569620, "full_name": "Tony Goss", "title": "Founder", "founder_bio": "cto @ idler. I (ethically) hacked Dark Forest to win $400k, beat the #1 ranked Dominion player, and play pro ultimate frisbee.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/tony-goss-a53732132", "x_url": "https://x.com/cha0sg0d_", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/536b6f75fb259031e056fa17525fc5aa24897b83.jpg" } ], "jobs": [ { "id": 106607, "title": "Forward Deployed Engineer", "url": "https://www.ycombinator.com/companies/idler/jobs/6bxHBTm-forward-deployed-engineer", "location": "San Francisco", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$125K - $400K", "equity_range": "0.01% - 1.00%", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 106667, "title": "Design Engineer", "url": "https://www.ycombinator.com/companies/idler/jobs/Lxnn9vi-design-engineer", "location": "San Francisco", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$125K - $300K", "equity_range": "0.01% - 1.00%", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 106668, "title": "Research Scientist", "url": "https://www.ycombinator.com/companies/idler/jobs/k8SRWwY-research-scientist", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$200K - $500K", "equity_range": "0.50% - 1.50%", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 106670, "title": "Software Engineer", "url": "https://www.ycombinator.com/companies/idler/jobs/D473KZp-software-engineer", "location": "San Francisco", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$125K - $400K", "equity_range": "0.01% - 1.00%", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 108287, "title": "Special Projects Operator", "url": "https://www.ycombinator.com/companies/idler/jobs/zTROCBy-special-projects-operator", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$125K - $400K", "equity_range": "0.10% - 1.00%", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 114955, "title": "Executive Assistant", "url": "https://www.ycombinator.com/companies/idler/jobs/nk4V3PX-executive-assistant", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$130K - $180K", "equity_range": "0.05% - 0.50%", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] }, { "id": 116440, "title": "Offensive Security RL Researcher", "url": "https://www.ycombinator.com/companies/idler/jobs/BpzzPnW-offensive-security-rl-researcher", "location": "San Francisco", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$200K - $500K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [] } ], "news": [], "launches": [ { "id": 111319, "title": "idler — push the frontier of your model", "tagline": "Hill-climb any benchmark, for any model, in any domain using our RL environments.", "body": "Today, we're introducing [Idler](https://idler.ai/) – a frontier data research lab. We build the evals and environments that the world's leading frontier labs use to measure and train their models.\n\nIn the last year, we’ve developed thousands of coding evals and environments for a top lab, continuously improving our product to keep pace with the frontier – even as models have gotten 8x better at long-horizon coding.\n\nNow, we're offering off-the-shelf datasets across an even broader set of capabilities: long-horizon software engineering, cybersecurity, recursive self-improvement, legal, long-horizon strategy & ops, enterprise safety, and many others.\n\n[ShelfLife](http://idler.ai/collections/shelflife), our first public benchmark, is out today — a digital twin seeded with the complete operational data of a live, profitable, growing multi-brand retailer running 9 stores, 3 commerce platforms, and a multi-supplier sourcing operation. Post-training Nemotron 3 Nano on 41 ShelfLife tasks raised its Finance Agent Benchmark score from 30.0% to 41.8% (p ≤ 0.001); full results and sample tasks are available [upon request](https://idler.cal.com/d/hpkJFogs2hcjrdqiSW9kCj/30min).\n\n**Our ask. Reach out if:**\n\n* You or someone you know is working on post-training at a lab or enterprise.\n* You have 10-100 employees and you’d like to supplement your revenue by licensing your codebase or de-identified business data.\n* You know any exceptional researchers, infra engineers, or ops people looking for their next challenge. We’re hiring!\n\nWe’d love to hear from you at [hello@idler.ai](mailto:hello@idler.ai). Thanks for the support!\n\n\\-[Nalu](https://x.com/naluconcepcion), [Ivan](https://x.com/chubivan?lang=en), & [Tony](https://x.com/cha0sg0d_).\n\n![uploaded image](/media/?type=post&id=111319&key=user_uploads/557067/a5fd0af4-18b9-449d-a871-63818866898d)\n\n", "url": "https://www.ycombinator.com/launches/SxT-idler-push-the-frontier-of-your-model", "votes": 12, "created_at": "2026-08-19T13:16:53.490Z" } ] }, { "id": 30851, "name": "Sylvian", "slug": "sylvian", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/9ce6100b0248b0cf8f017986931a26cfe36046f3.png", "website": "https://sylvian.ai", "all_locations": "San Francisco, CA, USA", "long_description": "Sylvian helps companies build specialized AI agents. ", "one_liner": "Specialized AI Agents.", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1762195938, "tags": [ "Reinforcement Learning", "B2B", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Fall 2025", "status": "Inactive", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/sylvian", "api": "https://devasheeshg.github.io/yc-api/batches/fall-2025/sylvian.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/sylvian", "x_url": "https://x.com/TheSylvianAI", "fb_url": null, "cb_url": null, "github_url": "https://github.com/SylvianAI", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/9ce6100b0248b0cf8f017986931a26cfe36046f3.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Aaron Epstein", "url": "https://www.ycombinator.com/people/aaron-epstein" }, "founders": [ { "user_id": 1252639, "full_name": "William Huang", "title": "Founder", "founder_bio": "CEO @ Sylvian. IPhO Gold, Ex-Waymo, Citadel Securities, Stanford CS.", "is_active": false, "linkedin_url": "https://www.linkedin.com/in/william-s-huang/", "x_url": "https://x.com/amiwilliamhuang", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/6a8e42f12ffb02233ed03f97d860d7a702b7bbcf.jpg" }, { "user_id": 2950012, "full_name": "Niall Kehoe", "title": "Founder", "founder_bio": "CTO @ Sylvian. Ex-Stanford CS, Two Sigma, Harvard Medical School / Stanford research", "is_active": false, "linkedin_url": "https://www.linkedin.com/in/niallkehoe", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/ec1c649325677c65811a0752c0e15aac2e351634.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 95160, "title": "Sylvian: Data for LLMs through Competition", "tagline": "Sylvian incentivizes high quality expert data for LLMs through Competition", "body": "We are [William Huang](https://linkedin.com/in/william-s-huang) and [Niall Kehoe](https://www.linkedin.com/in/niallkehoe/), and we’re launching [Sylvian](https://sylvian.ai)! We gather expert data for LLMs through competitions, starting with tool use (e.g. Excel, VSCode).\n\n![uploaded image](/media/?type=post&id=95160&key=user_uploads/1252639/cb059d06-59e0-402c-a49e-02752bbb9e17)\n\n**Problem**\n\nScaling laws dictate that LLMs continue to need lots of expert data, despite advances in RL. Even RL environments require data—for example, an environment that allows LLMs to edit spreadsheets in Excel would still require sample spreadsheets to define the task at hand.\n\nUnfortunately, existing data vendors like Scale or Mercor are not motivating the best experts through part-time pay.\n\n**Our Solution**\n\nSylvian hosts competitions where the best experts are motivated by the thrill of moving up leaderboards and the prestige of winning a large competition. We’re starting with tool use data because there are many expert communities surrounding tools like Excel and VSCode.\\\n\\\nWe already have 4,500+ experts, from IMO Golds to MIT/Stanford PhDs to full time QRs at Point72, producing data at 1B tokens/week!\n\nOur data is consistently at the frontier. Below is the result of one of our latest benchmarks we made with VSCode data from Data Science experts. See [sylvian.ai](https://www.sylvian.ai/blog/quant-benchmark) for more details.\n\n![uploaded image](/media/?type=post&id=95160&key=user_uploads/1252639/b874f47f-fef0-46c4-b26c-0d60887aadc5)\n\n**Our Story**\n\nWilliam won an IPhO Gold at 17 and Niall has won international coding contests at 13, since then we’ve went on to be a part of Stanford CS, Harvard Medical School, Citadel Securities, Two Sigma, and Waymo. \n\n**Our Ask**\n\nIf you’re in need of expert tool use data, we’d love to speak to you! You can reach us at [founders@sylvian.ai](mailto:founders@sylvian.ai) or visit [sylvian.ai](https://sylvian.ai).", "url": "https://www.ycombinator.com/launches/Okq-sylvian-data-for-llms-through-competition", "votes": 36, "created_at": "2025-11-04T19:46:52.391Z" } ] }, { "id": 30883, "name": "hillclimb", "slug": "hillclimb", "former_names": [ "Plun", "Hillclimb" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/660320374818f816f5cf14de03fe92dcab2740eb.png", "website": "https://hillclimb.com", "all_locations": "San Francisco, CA, USA", "long_description": "We work with frontier AI labs to help train their agents to become AI research scientists", "one_liner": "Training Data for Recursive Self-Improvement", "team_size": 4, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1759385382, "tags": [ "Reinforcement Learning" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Fall 2025", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/hillclimb", "api": "https://devasheeshg.github.io/yc-api/batches/fall-2025/hillclimb.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/hillclimb", "x_url": "https://x.com/hillclimbai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/660320374818f816f5cf14de03fe92dcab2740eb.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Garry Tan", "url": "https://www.ycombinator.com/people/garry-tan" }, "founders": [ { "user_id": 1934623, "full_name": "Jun Park", "title": "Founder/CEO", "founder_bio": "hillclimb / prev deepmind, pro valorant, georgia tech", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/jpark72/", "x_url": "https://x.com/jparkjmc", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/b2f60ab168221d0f4d4e9cd4fc7732880307a3a5.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 95533, "title": "hillclimb - Training data derived from human superintelligence", "tagline": "Building the human superintelligence community to advance AI", "body": "Hello! We’re Jun Park and Ibrakhim Ustelbay, founders of[ hillclimb](https://www.hillclimb.ing).\n\n\\_TLDR: To close the gap between today and the arrival of artificial superintelligence, we’re building a community of the world’s brightest minds to create the next generation of training data. Starting with math, we’ve formed the world’s densest cluster of IMO medalists, lean experts, and other math talent to design **RL environments** for frontier labs.\\_\\\n\\\n[Launch Video](https://youtu.be/30y3-SbRfZM)\n\n# The Problem - Data is accelerating on every axis\n\n**3 years ago,** we were impressed that ChatGPT could get good SAT scores. Today, GPT-5 and other frontier models assist meaningfully in coding, research, and tool use. Within two years, AI systems will autonomously [extend the frontier of science](https://x.com/SebastienBubeck/status/1958198661139009862?s=20).\n\n**Behind every leap in capability is a leap in the sophistication of data:**\n\n* From simple supervised fine-tuning → to richly structured RL environments → to world models and beyond\n* From mass labor → to curated, deeply intellectual data generation\n* From “labeling tasks” → to capturing the reasoning processes of domain experts\n\n![uploaded image](/media/?type=post&id=95533&key=user_uploads/1934623/128c37ee-f8e8-411b-9826-4b1b4492a9e2)\n\n_If AI reflects the intelligence of the people who shape its training data, the next era of progress requires the smartest humans on the planet._\n\n# The Solution - The world’s first _human superintelligence_ community\n\n![uploaded image](/media/?type=post&id=95533&key=user_uploads/1934623/1e971272-cd2e-46d9-9814-f81268047155)\n\n**In our first 2 months, we’ve:**\n\n* Built the densest cluster of elite math talent in the industry (IMO medalists, [Putnam Top 50](https://maa.org/wp-content/uploads/2025/03/2024-William-Lowell-Putnam-Competition-Announcement-of-Winners.pdf), Lean formalization experts, etc.)\n* Created _training data_ pipelines that faithfully capture their problem-solving intelligence\n* Partnered with [Nous Research](https://nousresearch.com) to help train a model that’s SOTA for math competitions.\n\n**We recruit top talent with a 85% winrate** against incumbents because we treat them as partners, not gig workers. For our work with Nous, we’ve negotiated co-authorship for our mathematicians because the data they produce has become a critical component of AI research.\n\n**Researchers love working with us** because we drop them in our community, allowing the iteration loop to become radically tighter. Instead of waiting for dataset revisions, they collaborate directly with world-class minds and see improvements in minutes, not months.\n\n# Our Story  **🎮 → 👨‍💻 → ⛰️**\n\nIt was only two years ago that we retired from competing professionally in esports. We taught ourselves to code and within a year, I became a research engineer at **DeepMind**, and Ibrakhim founded **Headstarter**, the first AI interviewer, scaling it to 100k+ users.\n\nIn the business of connecting the world’s smartest people to frontier labs, I understand the AI researcher’s perspective on what makes good training data, and Ibrakhim understands how to properly vet talent.\n\n![uploaded image](/media/?type=post&id=95533&key=user_uploads/1934623/08897fc4-b9c0-4dfd-913f-61f4bebd1998)\n\n# Our Ask\n\nIf you work at a frontier lab and need data (or know anyone who does), please reach out! Email us at [founders@hillclimb.com](mailto:founders@hillclimb.com).\n\nAlso, check out our site at [hillclimb.com](https://www.hillclimb.com/). Thanks for the support!\n\n![uploaded image](/media/?type=post&id=95533&key=user_uploads/1934623/797938d1-6d99-44f5-ad7d-1d302f0193fe)\n\n", "url": "https://www.ycombinator.com/launches/Oqr-hillclimb-training-data-derived-from-human-superintelligence", "votes": 155, "created_at": "2025-11-15T03:38:57.742Z" } ] }, { "id": 30930, "name": "Cortex AI", "slug": "cortex-ai", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/1c9246b8eba5ba12a7373a930ea46cc687dddb98.png", "website": "https://cortexrobot.ai/", "all_locations": "San Francisco, CA, USA", "long_description": "Cortex AI builds the world’s most diverse and large-scale real-world workplace robot & egocentric dataset — where the physical world becomes the next training and evaluation set for embodied AI.\r\n\r\nWe power frontier labs developing robotics foundation models and general-purpose robots by providing the data they need:\r\n1️⃣ Egocentric Data — real-workplace human video with hand/body pose, depth, and subtask labels.\r\n2️⃣ Robot Data — trajectories collected from manipulators and humanoids in real industry settings.\r\n3️⃣ Human-in-the-Loop Rollouts & Evals — real-world deployments with remote operators who recover robots when they fail, capturing data that feeds back into training and continuously improves models.\r\n\r\nAdditionally, through the Cortex Marketplace, workplaces get paid to host data-collection and evaluation sessions, while labs access the in-the-wild data that truly matters. This draws on Lucas’s previous experience as co-founder of Carousell, a C2C marketplace that scaled to a $1B+ valuation.", "one_liner": "Large-scale real-world robot & human data for embodied AI", "team_size": 3, "industry": "Industrials", "subindustry": "Industrials -> Manufacturing and Robotics", "launched_at": 1760045430, "tags": [ "Reinforcement Learning", "Robotics", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Fall 2025", "status": "Active", "industries": [ "Industrials", "Manufacturing and Robotics" ], "regions": [ "United States of America", "America / Canada", "Remote", "Partly Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/cortex-ai", "api": "https://devasheeshg.github.io/yc-api/batches/fall-2025/cortex-ai.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/cortexairobot", "x_url": "https://x.com/cortexairobot", "fb_url": null, "cb_url": null, "github_url": "https://github.com/cortexairobot", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/1c9246b8eba5ba12a7373a930ea46cc687dddb98.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Garry Tan", "url": "https://www.ycombinator.com/people/garry-tan" }, "founders": [ { "user_id": 53734, "full_name": "Lucas Ngoo", "title": "Founder/CEO", "founder_bio": "Founder and CEO of Cortex AI. We're building the world's most diverse real-world, real-workplace, and industry-scale egocentric + robot dataset. Prior to Cortex AI, Lucas was the co-founder and CTO of Carousell (valued at $1B+).", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/lucasngoo", "x_url": "https://x.com/lucasngoo", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/1b1f0f258b2fb62b1c7c32ca14e38ec0350b7ff4.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 95318, "title": "🦾 Cortex AI — Large-Scale Real-World Data for General-Purpose Robotics", "tagline": "Turning the real physical world into the next dataset for general-purpose robotics.", "body": "We’re building the world’s most diverse _real-world, real-workplace_ egocentric + robot dataset — where the **real physical world itself** becomes the next training and eval set for embodied AI.\\\n\\\n🌍 **The Problem**\\\nMost robotics datasets today are captured in lab environments or simulated settings — far removed from the **real physical world** where robots actually need to operate.\\\n\\\nThese datasets helped build the foundation models we have today, but the **next phase** of embodied AI needs to learn directly from the real world — from the noise, diversity, and complexity of real workplaces.\\\n\\\n🤖 **The Solution**\\\n\\\nFrontier labs use **Cortex AI** through 3 stages:\\\n1️⃣ **Get real-world egocentric data** → for world model + robotics foundation model pretraining\\\n2️⃣ **Collect real robot data** in real workplaces\\\n3️⃣ **Roll out policies** across real-world environments → the best evals happen in the real world, with our operators providing human-in-the-loop support to capture recovery data\\\n\\\nhttps://youtu.be/HD-weWSzy4Y\n\n\\\n🏭 **Cortex Marketplace**\\\nWe’re also launching the **Cortex Marketplace**, where _real workplaces get paid_ to host egocentric-capture and robot-teleop sessions — fueling a scalable pipeline of high-quality, real-world robotics data.\\\n\\\n👥 **The Team**\\\nPrior to Cortex AI, **Lucas Ngoo** co-founded and served as CTO of Carousell, scaling it from an idea into a $1B+ valuation marketplace used across Asia.\\\n\\\n🙏 **Our Ask**\\\nIf you work with or know teams at **frontier robotics labs** (humanoid companies, robotics foundation model research groups and teams), please make an intro ([lucas@cortexrobot.ai](mailto:lucas@cortexrobot.ai)) — we’d love to collaborate, thank you!\n\n🔗 [cortexrobot.ai](https://cortexrobot.ai)\\\n📩 [lucas@cortexrobot.ai](mailto:lucas@cortexrobot.ai)\n\n![uploaded image](/media/?type=post&id=95318&key=user_uploads/53734/069d07db-9cdf-4720-96b4-a7077818cb06)\n\n", "url": "https://www.ycombinator.com/launches/OnO-cortex-ai-large-scale-real-world-data-for-general-purpose-robotics", "votes": 20, "created_at": "2025-11-08T15:29:09.944Z" } ] }, { "id": 31139, "name": "Ashr", "slug": "ashr", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/36b682fdfb87515f8ca423d8a006eaac7229e134.png", "website": "https://ashr.io", "all_locations": "San Francisco, CA, USA", "long_description": "Ashr Manifold is a self-contained model training, hosting, observability, and continual learning platform for purpose-built open-weight models. \r\n\r\nStartups can use Manifold to post-train and manage fleets of continuously improving open-weight models without having a dedicated research team or infrastructure. Ultimately, this reduces operating costs and allows a greater degree of control over proprietary company knowledge.\r\n\r\nEnterprises can use Manifold to encode institutional judgements and priorities into open-weight models, reduce token costs, and manage every level of the AI life cycle, from data curation, to model serving on weights and infrastructure they own.", "one_liner": "Enterprise post-training, monitoring, and continual learning platform", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1771535371, "tags": [ "Deep Learning", "Generative AI", "Reinforcement Learning", "Data Engineering", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/ashr", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2026/ashr.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/ashr-io", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/36b682fdfb87515f8ca423d8a006eaac7229e134.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Harshita Arora", "url": "https://www.ycombinator.com/people/harshita-arora" }, "founders": [ { "user_id": 1467144, "full_name": "Rohan Kulkarni", "title": "Founder/CEO", "founder_bio": "Building custom models at Ashr Labs. Co-founder and CTO of Ask Geri (Acquired). Berkeley EECs. ", "is_active": true, "linkedin_url": "https://linkedin.com/in/rohankulkz", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/297857e249e4e8e3f4bce0f12fda39d67b33b506.jpg" }, { "user_id": 2584088, "full_name": "Shreyas Kaps", "title": "Founder", "founder_bio": "Building Ashr Manifold to help teams build and host purpose-built and trained models", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/shreyas-kaps-5b1a4420b/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/697406e5dc0beff2a338d68874683d7c8e1703c0.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 98286, "title": "Ashr: Mimic Your Production Environment and Users to Catch Agent Fails", "tagline": "Custom Tests and Evals Simulating Real User Behavior", "body": "\nHi YC, we’re Shreyas and Rohan — two ex-Berkeley students building Ashr.\n\n**TL;DR :** Ashr catches bugs and validates agent response quality by mimicking the patterns and quirks of real users. Developers today struggle to simulate these conditions in tests, resulting in real-time fails and unexpected behavior. We integrate as an SDK within your codebase.\n\n\n\n# The Problem\n\nAI automations are top of mind for every developer: large companies are integrating AI into older systems to stay technologically relevant, while AI-native companies are quickly emerging in every vertical. Unfortunately, most of these AI tools miss the mark: constant hallucinations, unexpected behavior, and dubious quality have eroded trust in agentic systems. In most cases, this is not because foundation models aren’t powerful enough, but because of subtle prompting, tooling, and scaffolding errors that only emerge with rigorous testing. The need for high volumes of realistic testing arises in many scenarios: we’ve spoken to companies who want to prove that their agents are capable of handling sensitive data to potential customers, developers who want to make their agents work better, and engineers who want to test edge-cases and new features but don’t have all the necessary tools.\n\nWe’ve seen teams confront this problem with a litany of half-measures: pushing to production and tanking user trust as a means of discovering bugs, half-baked test harnesses, and vibe-coded test suites that only scratch the surface. We built ashr to solve this.\n\n# Our Solution\n\nAshr is a modular testing and evaluation platform for your agentic system. We integrate as an SDK into your codebase, allowing you to request test suites and analyzing your agents’ output based off your tooling and ideal outputs. These tests mimic how real users utilize an agentic system: taking complex tool call paths, using obscure language with unclear instructions, and attempting to use restricted features. We develop environments, fine-tuned custom scorers, and stop at nothing to provide all the insights needed to bring your agent to a truly production-ready state.\n\nWe also work seamlessly with Claude Code and Cursor, allowing your coding agents to quickly prototype and test changes while receiving immediate feedback. Teams have run swarms of agents testing with ashr overnight, and have woken up to changes that reduced hallucinations and increased accuracy and user retention, all without writing a single line of code.\n\n# Our Ask\n\nIf you’re building agents, book a call today at [**https://ashr.io/**](https://ashr.io/). **All onboarding is done via call so we can ensure everyone is ready to go and start using as soon as possible.** In addition, we’d love feedback on our docs, opinions about the nature of test infra, and just chatting with a bunch of builders on how to make our product better.\n\nYou can reach us at [founders@ashr.io](mailto:founders@ashr.io)\n\n# The Team\n\n**We’re two young founders passionate about changing the world with agentic development**\n\n* **Rohan** built his first startup, an AI-native survey platform for old age homes, as a student at Berkeley, before exiting as part of an acquisition and joining YC.\n* **Shreyas** worked on AI-native workflows at Tessera Labs, mentored by a team of ex-Meta and Netflix research, and delivering value in the ERP transformations of Fortune 100 companies.", "url": "https://www.ycombinator.com/launches/PZG-ashr-mimic-your-production-environment-and-users-to-catch-agent-fails", "votes": 9, "created_at": "2026-02-27T22:43:17.702Z" } ] }, { "id": 31181, "name": "Synthetic Sciences", "slug": "synthetic-sciences", "former_names": [ "InkVell" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/d12b5cddde967f28e71a5bdfa173bbd8e7a0b10c.png", "website": "https://syntheticsciences.ai/", "all_locations": "San Francisco, CA, USA", "long_description": "Synthetic Sciences is a research lab building infrastructure for scientific superintelligence. Empirical science is bottlenecked by verification that is slow, expensive, and imperfect. We work on tooling, evals, training environments, algorithms, and labs that make autonomous science possible.", "one_liner": "Infrastructure for Autonomous Science", "team_size": 4, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1770489645, "tags": [ "Reinforcement Learning" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2026", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/synthetic-sciences", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2026/synthetic-sciences.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/synsci", "x_url": "https://x.com/SynScience", "fb_url": null, "cb_url": null, "github_url": "https://github.com/synthetic-sciences", "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/d12b5cddde967f28e71a5bdfa173bbd8e7a0b10c.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Gustaf Alstromer", "url": "https://www.ycombinator.com/people/gustaf-alstromer" }, "founders": [ { "user_id": 1441373, "full_name": "Ishaan Gangwani", "title": "Co-Founder", "founder_bio": "Prev: IOAI '25 (HM), USACO Platinum, Z-Fellows, Emergent Ventures", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ishaan-gangwani-3a7792241/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e1be13bfebc500d9569596d86efa347dd69024d0.jpg" }, { "user_id": 2581686, "full_name": "Aayam Bansal", "title": "Co-Founder", "founder_bio": "Built COVID-19 helpline infrastructure for Indian govt serving 50k+ people daily\nPatented (US, SG, IND) and Sold Orthopaedic AISock for 5-figures at 17\nPublished at NeurIPS, ICML, ICLR, AAAI, CVPR & other A* conferences\n\nPrev: aisock (Acq.), Z-Fellows, Emergent Ventures", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/aayambansal/", "x_url": "https://x.com/aayambansal", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/a8d0df5f64e75745de4e4b1b943ddbf7f0459d29.jpg" } ], "jobs": [], "news": [ { "title": "Indian teens among youngest founders to get accepted into Y Combinator W26 - The Economic Times", "url": "https://economictimes.indiatimes.com/ai/ai-insights/indian-teens-among-youngest-founders-to-get-accepted-into-y-combinator-w26/articleshow/125788177.cms?from=mdr", "date": "Dec 08, 2025" } ], "launches": [ { "id": 98217, "title": "Synthetic Sciences – AI Co-Scientists for End-to-End Scientific Research", "tagline": "Cursor for Scientific Research.", "body": "Hi! We’re Aayam and Ishaan, co-founders of [Synthetic Sciences](https://syntheticsciences.ai/).\n\n# **TL;DR**\n\nSynthetic Sciences is a platform where scientists delegate complex research tasks to AI co-scientists that handle the full research loop: **literature → hypotheses → experiments → results → paper drafts**.\n\nOver the last year, AI has been showing some traction in math research. Math gets the coverage because it’s easy to measure. But there is a non-trivial unlock to be made in accelerating the rest of science – ML research, computational biology, proteomics - where the tooling hasn't caught up yet.\n\n**Launch Video:** \n\n# **The Problem**\n\nScientific research is a marathon of iterations. Read hundreds of papers. Form hypotheses. Design and run experiments. Wait. Analyze. Write. Revise. Submit. Each cycle compounds.\n\nThe existing tooling doesn't help. AI “literature review” tools surface papers but don't carry context forward into your code or experiments. Jupyter, W&B & others track runs but are siloed from your literature and writing. Writing happens in isolation, even though it could be steering decisions earlier.\n\nResearch is a long chain of context-dependent steps, and our tools treat each one like it’s isolated. Every iteration forces you to rebuild state: what you read, what you tried, why you chose this method, what failed, what changed.\n\nA single research question becomes months of fragmented work scattered across tools, tabs, and time zones. And it's all on you. Every step requires your attention, your context, your time.\n\nYou can't parallelize yourself. If we could accelerate scientific research the way we’ve accelerated software engineering in the last 24 months we are guaranteed to see science breakthroughs that will massively benefit humanity.\n\n# **Our Solution**\n\nPoint synsci at a research question, a repo, or a dataset. Our AI co-scientists autonomously:\n\n* Search + synthesize literature grounded in your project\n* Generate high-quality hypotheses + idea trees tied to prior work\n* Design experiment plans + GPU job specs\n* Write and run code (Python, R, ML pipelines) in containerized environments\n* Launch and monitor experiments on serverless GPUs & GPU clusters\n* Output publication-ready drafts: LaTeX, figures, slides\n\nWe’ve mostly tested synsci on ML research so far, but we’re rapidly expanding across domains. It’s already strong in computational biology research: on BixBench Verified, our biology mode achieves state-of-the-art performance (92%).\n\n![uploaded image](/media/?type=post&id=98217&key=user_uploads/1441373/0d71bacf-1ac6-4ee5-973e-42a543973d04)\n\n# **Flywheel Mode: Own Your Models**\n\nIf you have users, you already have the rarest ingredient in AI: proprietary signal. Every correction, accept/reject, trace, and outcome becomes training data. Everyone has access to the same models and can buy the same tokens. Nobody else can buy your users, your logs, and your feedback loops. \n\nStrong open-weight models now make it possible to beat frontier performance on your specific tasks through post-training (SFT + RL). However, the problem is everything around the model: data plumbing, evals, deployment, and compute. That friction keeps most teams renting intelligence by the token instead of owning it.\n\nFlywheel Mode makes the loop painless. We handle the glue code and ship the workflow end-to-end across 20+ compute providers and hundreds of built-in skills. Use your AI co-scientists to run the full loop: train, evaluate, deploy, iterate.\n\n![uploaded image](/media/?type=post&id=98217&key=user_uploads/1441373/ea6a1549-dd1f-40e3-bd36-ae627be085b0)\n\n# **Our Story**\n\nWe met doing ML research at NUS, CMU and MIT CSAIL & published together at NeurIPS, ICML, and AAAI workshops. Ishaan came from competitive programming (top-ranked in India for IOAI Team Selection, USACO Platinum). Aayam previously built, patented & sold an AI orthopaedic sock and built COVID-19 infra serving 50k+ users daily. \n\n# **Our Ask**\n\n* **Try it:**[ ](https://app.syntheticsciences.ai)[syntheticsciences.ai](http://syntheticsciences.ai) - $5 in free credits when you sign up!\n* **If you’re a CTO / Head of R&D / research lead** and want to accelerate R&D with AI co-scientists, we’d love to talk.\n* **RL environments for scientific research:** We're building LHRL environments to post-train models on ML research workflows. If you're an AI lab interested in process-oriented data, let's talk.", "url": "https://www.ycombinator.com/launches/PY9-synthetic-sciences-ai-co-scientists-for-end-to-end-scientific-research", "votes": 20, "created_at": "2026-02-26T18:00:26.592Z" } ] }, { "id": 31191, "name": "Traverse", "slug": "traverse", "former_names": [ "Clice AI", "Clice" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/35a08dba94f5c7c4efe9e2c566eb21cc8f85b0e1.png", "website": "https://www.traverse.so", "all_locations": "San Francisco, CA, USA", "long_description": "", "one_liner": "Custom models and inference, built and run for enterprises", "team_size": 1, "industry": "B2B", "subindustry": "B2B", "launched_at": 1767260558, "tags": [ "Machine Learning", "Reinforcement Learning", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/traverse", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2026/traverse.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/traverse-so/", "x_url": "https://x.com/traverseso", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/35a08dba94f5c7c4efe9e2c566eb21cc8f85b0e1.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jared Friedman", "url": "https://www.ycombinator.com/people/jared-friedman" }, "founders": [ { "user_id": 2775848, "full_name": "Lance Yan", "title": "Founder/CEO", "founder_bio": "Seeking truth.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/lance-yan/", "x_url": "https://x.com/lanceyyan", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/19d6b3aa8f7c40b89a2db639a278aa326cc3a1f5.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 98535, "title": "Traverse - Data Research for the Non-Verifiable", "tagline": "We give AI labs the training data that enables their frontier models to acquire taste", "body": "**TLDR:** Traverse is a data research lab focused on the hardest unsolved problem in frontier model training: **ambiguous, taste-dependent work**. \n\nWe capture real human behavior in highly controlled settings and turn judgment, taste, and long-horizon reasoning into high-fidelity environments for frontier models.\n\n\n\n**The Problem**\n\nReinforcement learning has already helped frontier models achieve superhuman ability in math and coding since these domains have very clean and deterministic feedback loops. \n\nHowever, the next frontier domains (law, healthcare, sales, writing, etc) aren’t as black and white. Current approaches fail to accurately represent subjective scenarios, despite these domains being where most of the economic value is.\n\n**The Solution**\n\nWe are building data factories to create high fidelity real life work environments at scale.\n\nTraverse captures what people actually think while doing their work; we are pioneering a new category of training data that is yet to exist.\\\n\\\n**If you’d like to understand the details, reach us at**\\\n[lance@traverse.so](mailto:lance@traverse.so)\\\n\n\n![uploaded image](/media/?type=post&id=98535&key=user_uploads/2775848/4586ce6d-bd41-40af-9be6-586ab6226d02)\n\n", "url": "https://www.ycombinator.com/launches/PdH-traverse-data-research-for-the-non-verifiable", "votes": 41, "created_at": "2026-03-06T16:12:51.971Z" } ] }, { "id": 31231, "name": "Anchorhead", "slug": "anchorhead", "former_names": [ "Willow", "Glue" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/7b4d1a0f867d8f5d24359da3d4d5fca3fbbabe57.png", "website": "https://www.anchorhead.ai", "all_locations": "San Francisco, CA, USA", "long_description": "We build custom evals for product teams, as a service. To build these eval sets, we source domain experts and have them turn their work into original, real world evals.\r\n\r\nFor example, our research engineering experts have create tasks involving optimizing an algorithm, deploying a model, or running experiments to solve a novel problem. A grader scores the agent’s performance, and these scores serve as signals during evaluations.", "one_liner": "Evals for frontier research agents", "team_size": 1, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1772831162, "tags": [ "Reinforcement Learning" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Winter 2026", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/anchorhead", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2026/anchorhead.json", "year_founded": 2025, "city": "San Francisco", "country": "US", "linkedin_url": "", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/7b4d1a0f867d8f5d24359da3d4d5fca3fbbabe57.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jon Xu", "url": "https://www.ycombinator.com/people/jon-xu" }, "founders": [ { "user_id": 965938, "full_name": "Perbhat Kumar", "title": "Founder/CEO", "founder_bio": "ceo at Anchorhead. prev engineer at Amazon and Microsoft. worked on a home launcher redesign for Kindle Fire, and built enterprise-scale rag agents at Microsoft Copilot Studio", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/perbhat", "x_url": "https://www.x.com/perbhatk", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/4ca329a4553dab624f6d6ee0231989b751799739.jpg" } ], "jobs": [], "news": [], "launches": [] }, { "id": 31285, "name": "Brumby (Formerly GrazeMate)", "slug": "brumby", "former_names": [ "GrazeMate" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/e7f64c3261f306e1859d503d40ad84601c127c82.png", "website": "https://brumby", "all_locations": "Sydney, NSW, Australia", "long_description": "Brumby builds autonomous drones that herd cattle.\r\n\r\nOn command, our drones fly to a paddock, position themselves around the mob, and move them where they need to go. What used to take a full day of helicopters, motorbikes, and horses now runs on a schedule.\r\n\r\nWe work with some of the largest cattle ranches in the world. While the drones are herding, they're also estimating animal weights, measuring grass biomass, monitoring water levels, and flagging sick animals. We're building physical AI that lets a grazier manage thousands of head across millions of acres from their phone.", "one_liner": "Robot Cowboys that Herd Cattle with AI Drones", "team_size": 3, "industry": "Industrials", "subindustry": "Industrials -> Agriculture", "launched_at": 1768433444, "tags": [ "Reinforcement Learning", "Drones", "Computer Vision", "Agriculture" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Winter 2026", "status": "Active", "industries": [ "Industrials", "Agriculture" ], "regions": [ "Australia", "Oceania", "Remote", "Partly Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/brumby", "api": "https://devasheeshg.github.io/yc-api/batches/winter-2026/brumby.json", "year_founded": 2025, "city": "Sydney", "country": "AU", "linkedin_url": "https://www.linkedin.com/company/105884860", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/e7f64c3261f306e1859d503d40ad84601c127c82.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Tyler Bosmeny", "url": "https://www.ycombinator.com/people/tyler-bosmeny" }, "founders": [ { "user_id": 2933487, "full_name": "Sam Rogers", "title": "Founder/CEO", "founder_bio": "CEO of Brumby.\n\nSpent half my life on a cattle farm, and the other half building robots.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/samson-rogers/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/9019d14f0cd67dacb9f12edba5539dbeb3470e9f.jpg" } ], "jobs": [ { "id": 112676, "title": "Remote Pilot", "url": "https://www.ycombinator.com/companies/brumby/jobs/4L3nppN-remote-pilot", "location": "Sydney", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$85K - $150K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 112677, "title": "Embedded Systems Engineer", "url": "https://www.ycombinator.com/companies/brumby/jobs/d4z6kbg-embedded-systems-engineer", "location": "Sydney", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $150K", "equity_range": "", "experience": "3+ years", "visa": "US citizenship/visa not required", "skills": [ "Bash/Shell", "C++", "Microcontrollers", "PCB design" ] }, { "id": 112678, "title": "Operations Associate", "url": "https://www.ycombinator.com/companies/brumby/jobs/n8GBsUB-operations-associate", "location": "Sydney", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "$100K - $130K", "equity_range": "", "experience": "1+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 112679, "title": "Perception Engineer", "url": "https://www.ycombinator.com/companies/brumby/jobs/A0m4omK-perception-engineer", "location": "Sydney", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $150K", "equity_range": "", "experience": "3+ years", "visa": "US citizen/visa only", "skills": [ "C++", "Python", "Machine Learning", "Deep Learning", "OpenCV", "Computer Vision" ] }, { "id": 112680, "title": "Software Engineer", "url": "https://www.ycombinator.com/companies/brumby/jobs/ar7aJ98-software-engineer", "location": "Sydney", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "$100K - $160K", "equity_range": "", "experience": "Any (new grads ok)", "visa": "US citizen/visa only", "skills": [ "Amazon Web Services (AWS)", "iOS", "PostgreSQL", "Distributed Systems", "Terraform", "FastAPI" ] } ], "news": [ { "title": "GrazeMate Raises $1.2 Million As 19-Year-Old CEO Automates Cattle Mustering", "url": "https://www.forbes.com.au/news/innovation/grazemate-raises-1-2-million-19-year-old-ceo-cattle-mustering/", "date": "Jan 13, 2026" } ], "launches": [ { "id": 97388, "title": "GrazeMate (Now Brumby): Drones that Herd Cattle", "tagline": "We build autonomous “robot cowboys” that help ranchers move and monitor cattle.", "body": "**TL;DR** - Ranchers can spend hundreds of thousands per year moving cattle between fields. GrazeMate builds autonomous drone software to guide cattle - doing the work of helicopters and motorbikes with AI that understands animal behaviour.\n\n\n\n**❌ The problem**\n\nI'm Sam. I grew up on my family's cattle station in Australia, where we move cattle on motorbikes, helicopters, and horseback. It took hours each day and >$100k/year.\n\nIn the US, ranches spend anywhere from $10k–$1M per year and hours each day moving cattle just to keep them on fresh grass. It's expensive, risky, and hard to staff.\n\n**✅ The solution**\n\nGrazeMate's autonomous drones herd your cattle for you. What could take teams a full day now takes three clicks from a phone.\n\nOur system:\n\n* Finds your herd and guides them calmly to the next pasture\n* Uses AI that reads animal behaviour and applies pressure like an experienced stockman\n* Learns how your cattle respond and improves with every move\n\nWhy ranchers are switching:\n\n* More frequent moves → Better grazing, pasture use, and stocking rates\n* More affordable than traditional methods\n* Hours of herding → seconds on a phone\n* Real-time weight estimates, pasture data, and infrastructure insights\n\n![uploaded image](/media/?type=post&id=97388&key=user_uploads/2933487/a7e07817-bfde-4784-90ab-e4f9d16579b8)\n\n**🤠 Why us**\n\nMy family runs 6,000 head. I dropped out of robotics at University of Sydney after seeing both the challenges with modern ranching, and the capacity robotics has to help. Our team combines PhDs and engineers who grew up on ranches.\n\n**🙏 Our ask**\n\nWe're onboarding early partners. We’d be extremely grateful for introductions to:\n\n* Large grazing operations (US and beyond)\n* Ranch managers exploring rotational grazing\n* Industry leaders, cattle agents, feedlots, processors\n\nAnd if that’s you, schedule a demo on grazemate.com, or contact me at [sam@grazemate.com](mailto:sam@grazemate.com). I’d love to chat!", "url": "https://www.ycombinator.com/launches/PKm-grazemate-now-brumby-drones-that-herd-cattle", "votes": 30, "created_at": "2026-02-02T15:42:29.970Z" } ] }, { "id": 32130, "name": "Olam Labs", "slug": "olam-labs", "former_names": [ "Gud Labs" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/d6d9543a0527b0b5ac428e043f211c75ac5d1d59.png", "website": "https://olamlabs.ai", "all_locations": "San Francisco, CA, USA", "long_description": "We research complex, high-fidelity simulations for training and evaluating frontier AI models, alongside the infrastructure to test and deploy them at scale. Our research focuses on simulations with realistic socialization, multi-agent interaction, swarms, and ultra-long-horizon tasks.\r\n\r\nTry one of our first releases, Multi-Agent Arena (https://olamarena.com/), where humans come play social strategy games against multiple AI agents. It has multi-agent social games like Poker, Diplomacy, Risk, and more.", "one_liner": "Crafting simulations for frontier lab model training", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1784577901, "tags": [ "Reinforcement Learning", "Gaming", "Data Engineering", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": [ { "question": "How did your company get started? (i.e., How did the founders meet? How did you come up with the idea? How did you decide to be a founder?)", "answer": "Om and Shreshth met in middle school. They were the only two kids in the “gifted” program in their grade, and their first project together was running a snacks and candy store at their school where the profits were used to donate to charity.\\\r\n\\\r\nDuring Christmas holiday in 2025, Om created a multi-agent simulated environment where agents ran their own company in markets competing against each other, and showed it to Shreshth.\\\r\n\\\r\nShreshth, with a background in data and RL among other things, loved the implications of it. Om discussed with him how if multi-agent environments were developed and productionized at scale via frameworks, APIs for infra, and interactive games the kind of data and evaluations it could generate.\\\r\n\\\r\nThe problem was (and still is) that despite the success of multi-agent environments in research, the ecosystem around them has not been pursued to a degree where the utility of multi-agent environments is made readily available to everyone.\\\r\n\\\r\nThey decided to pursue it and applied to YC, and have been in pursuit of a future where we can understand and measure the models that we work with everyday and that will become increasingly autonomous in our everyday lives with AI acceleration. Om and Shreshth both believe there is a severe shortage of good AI evaluations, especially on qualitative behaviors, of which multi-agent environments (and allowing others to build them) is very well suited to solve.\\\r\n\\\r\nMore on Shreshth: [shresh.ca](http://shresh.ca) Shresh’s X: \\\r\nMore on Om: [sensho.xyz](https://www.sensho.xyz/) Om’s X: " } ], "url": "https://www.ycombinator.com/companies/olam-labs", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/olam-labs.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/olam-labs/", "x_url": "https://x.com/olam_labs", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/d6d9543a0527b0b5ac428e043f211c75ac5d1d59.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/3e94d77a846523415d54724a99c9965d17599d69.png" ], "primary_partner": { "name": "Harshita Arora", "url": "https://www.ycombinator.com/people/harshita-arora" }, "founders": [ { "user_id": 3676793, "full_name": "Om Buddhdev", "title": "Founder/CEO", "founder_bio": "Building simulated worlds for frontier labs to train capable and safe models.\n\nI love learning, tech, competing, and building cool things. I am optimistically AGI pilled and believe most people (even in SF) have not yet internalized what will change over the next 10Y.\n\nPreviously a Staff Engineer @ latitude.io, one of the best Valorant players in the world, math research cited by Terence Tao, TikTok creator /w 190k followers, Roblox game dev, and some other things.\n\nhttps://www.sensho.xyz", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ombud/", "x_url": "https://x.com/sensho", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/91f0defa43559eee579ccd34e4f7fd404cdb2aa6.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 109432, "title": "Can you beat frontier LLMs at social strategy games? | Multi-Agent Arena by Olam Labs, evaluating through multi-agent simulations", "tagline": "Play games like Risk, Poker, or Codenames against frontier AIs. We're running public long-horizon games to do evaluations on social behavior and agentic performance.", "body": "**TL;DR:** As many researchers and labs will say, there is a massive shortage of genuinely good AI evaluations. Multi-agent environments are an underdeveloped way of evaluating models that we think will help address our increasingly worse shortage of evals. We're starting with **Multi-Agent Arena**, where anyone can come compete against frontier AI models in social multiplayer games, as we build towards the full stack for multi-agent environments. Try Multi-Agent Arena for free at :) \\\n\\\n[https://drive.google.com/file/d/186GgmnOLz4Yc1BpFk6AxHd3aqUdFYCQy](https://drive.google.com/file/d/186GgmnOLz4Yc1BpFk6AxHd3aqUdFYCQy/view?usp=sharing)\n\n---\n\n## **Overview**\n\nMulti-agent environments provide an essential, underutilized framework for benchmarking autonomous AI models. Multi-Agent Arena is an interactive platform where human participants compete against frontier AI models in multiplayer games, serving as an experimental foundation to evaluate complex agent interactions. Multi-Agent Arena is the first of many stepping stones to building the entire stack and ecosystem for multi-agent environments.\\\n\\\n**The Evaluation Gap** \\\nCurrent AI evaluations primarily focus on single-agent, task-oriented capabilities. While benchmarks such as SWE-bench measure functional task execution, they do not assess the dynamic social behaviors required for autonomous real-world deployment: negotiation, coalition stability, long-horizon planning against adaptive opponents, and spontaneous deception.\n\nSimulated multi-agent environments induce these complex interactions while preserving complete experimental control. Unlike unstructured real-world deployment, these environments offer full data legibility, enabling researchers to inspect every reasoning trace, state transition, and stochastic variable behind model decisions.\n\n**Our Preliminary Evaluations**\\\nThe platform runs deterministic, replayable environments hosting human players alongside anonymized frontier models. Structured as accessible multiplayer games, the infrastructure captures interaction data to score models across two primary axes in our first evaluated game, Poker:\n\n* **Competitive Capability:** One of our preliminary methods is using a Bradley-Terry model to yield Elo ratings with bootstrapped confidence intervals, supplemented by domain-specific metrics such as big blinds per 100 hands in poker. Top-quartile human participants currently outperform all benchmarked models in Poker.\n* **Behavioral Diagnostics:** Model reasoning traces are evaluated against ground-truth game states using expert-curated rubrics and variance-checked grader panels to ensure inter-annotator agreement.\n\n**Initial Metric: Deception Index** \\\nThe benchmark's initial diagnostic metric, the Deception Index, quantifies deliberate misrepresentation per 10,000 turns. A deceptive event is logged only when an agent's private chain-of-thought confirms awareness of a true state while its external communication directly contradicts it. In baseline tests with no explicit instructions or prompts to deceive, frontier models demonstrated performance variance spanning nearly two orders of magnitude across identical environments.\n\nComplete evaluations, research methodology and technical documentation are available at [olamlabs.ai/research/social-arena](https://olamlabs.ai/evaluations)\\\n\\\n**Launch**\\\n\\\nWe already passed a thousand matches a day in our early access, are nearing a billion tokens used per day, ranking on the OpenRouter leaderboards, and have thousands of users that are paying to play in the arena while we work with researchers on use cases they have for our work and environments. We’re hoping this is just a stepping stone and want to scale towards our roadmap and ambitions of what we believe multi-agent environments, when properly developed with the right stakeholders in the training/eval ecosystem, can achieve for AI progress.\\\n\\\n\\\n**What’s in it for you?**\n\n1. **Play.** It's free, and every match directly informs the evals. Bring friends, try to beat the models, it’s harder than you'd think. 🃏\n2. **Speak with researchers.** We love speaking with researchers interested in this space and it’s all that we think about; if that’s you please feel free to reach out to us at [founders@olamlabs.ai](mailto:founders@olamlabs.ai) or on \\\n \\\n 🫡\n\n---\n\n## About the team\n\n**Om Buddhdev (CEO)** — Previously a Product Engineer then promoted to Staff Engineer at AI Dungeon / Latitude working on AI infra, agents, and consumer platform work, left for Olam Labs. Before that he was among the best Valorant players in the world (top 25, 8x radiant), at 17 a large content creator on TikTok with 195k followers, and as a young teenager made Roblox games played millions of times, among other things in his past. More on Om at \n\n**Shreshth Sharma (CTO)** — Background in production data engineering: was a Data Scientist at RBC while doing a double degree at the University of Waterloo in CS and Math, dropped out of both for Olam Labs. He has a background in many data and statistics projects, alongside briefly being a top 50 Fifa player. More on Shreshth at \\\n\\\nOm and Shreshth have been friends since middle school, and Olam Labs was born out of their personal interest and desire to build out the entire stack for multi-agent environments to be the default way we evaluate models in the future.", "url": "https://www.ycombinator.com/launches/ST2-can-you-beat-frontier-llms-at-social-strategy-games-multi-agent-arena-by-olam-labs-evaluating-through-multi-agent-simulations", "votes": 9, "created_at": "2026-08-05T21:32:50.566Z" } ] }, { "id": 32156, "name": "Enact", "slug": "enact", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/6f79f9bbc22f20d6e3465ca66081c4a646834064.png", "website": "https://enact.company", "all_locations": "San Francisco, CA, USA", "long_description": "Enact creates the physical environments to benchmark and hill climb robotics models on high-value tasks. We find where robots fail, and fix them with our in-house training loop.", "one_liner": "Real-world RL environments for robotics models", "team_size": 2, "industry": "Industrials", "subindustry": "Industrials -> Manufacturing and Robotics", "launched_at": 1786311090, "tags": [ "Reinforcement Learning", "Robotics", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "Industrials", "Manufacturing and Robotics" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/enact", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/enact.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/enact-company/?viewAsMember=true", "x_url": "https://x.com/enact_labs", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/6f79f9bbc22f20d6e3465ca66081c4a646834064.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jon Xu", "url": "https://www.ycombinator.com/people/jon-xu" }, "founders": [ { "user_id": 2406199, "full_name": "James Stevens", "title": "Founder/CEO", "founder_bio": "Founder & CEO @ Enact", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/james-stevens-440362200", "x_url": "https://x.com/jamesw_stevens", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/c605b46207ec2b29443db68751da93ca45c507c0.jpg" }, { "user_id": 3594727, "full_name": "Govind Chada", "title": "Founder/CTO", "founder_bio": "Founder at Enact. Previously: Robotics research in Chelsea Finn’s IRIS Lab at Stanford AI Lab, Meta Reality Labs.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/govind-chada", "x_url": "https://x.com/govindchada", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/bffdb6ffe1344f591c02771b2dceccd816fe305e.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 109880, "title": "Enact: post-training infrastructure for robotics models", "tagline": "Enact generates the data and evaluations needed to make robotics models reliable in the real world.", "body": "**TL;DR: Enact helps teams deploy robots that work in the real world. We roll out policies, generate failure / recovery data, and run in-house evaluations so robots can complete real tasks in production.** \n\n[**https://www.youtube.com/watch?v=dam0tQBIezY**](https://www.youtube.com/watch?v=dam0tQBIezY)\n\n**The Problem**\n\nRobotics models can look flawless in choreographed demos and still fail when an object slips, the scene shifts, or the robot makes an early mistake. Those failures push the policy into states its training data never taught it to recover from. **Robotics models must train on edge cases to be reliable in the real world.**\n\n**What We Built**\n\nEnact turns a policy’s failures into targeted recovery datasets for teams training robotics models. We use the data to fine-tune models and run real-world evaluations, verifying our data improves reliability.\n\n**We’re already serving our first customers: finding where their robots fail and collecting the targeted recovery data needed to make them robust.**\n\n* Roll out an existing model on a physical robot.\n* Identify recurring failure states.\n* Recreate those states and demonstrate how to recover.\n* Retrain with the new trajectories added in.\n* Repeat\n\n**Targeted recovery data pushes success rates above 99%.**\n\nOn our controlled packing task using pi0.5 on the YAM 6DoF arms:\n\n* 200 ordinary expert demonstrations: 90/100 successful physical rollouts\n* 250 ordinary expert demonstrations (more of the same data): 89/100\n* 200 expert demonstrations plus 50 targeted recovery demonstrations: 99/100\n\n![uploaded image](/media/?type=post&id=109880&key=user_uploads/2406199/e1b83c19-6e1d-4dda-adcf-258398bc1676)\n\n**We’re continuing to longer-horizon tasks such as industrial kitting, lab automation, and electronics recycling**\n\nLonger-horizon tasks need substantially more base and recovery data, but the loop is the same: find failures, target missing behaviors, retrain, repeat.\n\n**Who We Are**\n\nWe’re James Stevens and Govind Chada, and we met at Stanford. Govind researched real-world robot adaptation in Chelsea Finn’s IRIS Lab and co-developed recovery methods published at top robotics conferences. Together, we built Enact’s in-house data and evaluation operation and are already serving robotics teams. We started Enact to make robotics models work reliably outside the lab.\n\n**Our Ask**\n\nTraining a robot foundation model or deploying a policy that fails on a real task? Tell us the model, task, and where it breaks, and we’ll make it work.\n\nEmail: [founders@enact.company](mailto:founders@enact.company)", "url": "https://www.ycombinator.com/launches/SaG-enact-post-training-infrastructure-for-robotics-models", "votes": 16, "created_at": "2026-08-11T14:30:00.204Z" } ] }, { "id": 32198, "name": "Riften", "slug": "riften", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/d3c58cea1afce06f81a0352c8f3ecc8554d707c7.png", "website": "https://riften.ai", "all_locations": "San Francisco, CA, USA", "long_description": "Riften builds model routing, evaluations, private models, and training environments for AI labs, AI-native companies, and enterprises.\r\n\r\nWe help teams choose models for their workloads, evaluate how they perform, and build private models around company-specific work. We’re also developing coding environments where agents learn through backend engineering tasks.", "one_liner": "Earned intelligence for every company", "team_size": 5, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1785872995, "tags": [ "Developer Tools", "Reinforcement Learning", "Infrastructure", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/riften", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/riften.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/riftenai/", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/d3c58cea1afce06f81a0352c8f3ecc8554d707c7.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Harshita Arora", "url": "https://www.ycombinator.com/people/harshita-arora" }, "founders": [ { "user_id": 2941426, "full_name": "Carl Stauffer", "title": "Founder/CEO", "founder_bio": "Founder and CEO of Riften. Former Palantir Forward Deployed Engineer. Former nuclear researcher at Jefferson Lab. CS and AMS at Johns Hopkins.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/carl-stauffer-314335262/", "x_url": "https://x.com/carlstauffer_", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/0a3a2f44ecdd315423ad0b8510ef3bffafaf4843.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 110330, "title": "Riften: Route every AI request. Build the models your company owns.", "tagline": "Cut LLM spend with one endpoint that selects the lowest-cost model that works.", "body": "[Riften Launch Video](https://youtu.be/9T82XHysqsc)\n\n## What Riften does\n\nRiften is an OpenAI- and Anthropic-compatible gateway that routes each LLM request to the lowest-cost model capable of completing it.\n\nConnect your existing providers, change the base URL, and keep the product you already built. Riften works across customer-facing AI products and internal tools, including Claude Code and Codex.\n\n**Riften is already removing enterprise-scale model spend from production workloads.**\n\n## Why we built it\n\nMost AI products choose a model once and send nearly everything to it.\n\nA documentation lookup, a support classification, and a codebase migration do not require the same intelligence. Yet they are routinely sent to the same frontier model at the same price.\n\nFor customer-facing products, that cost cuts directly into gross margin. For internal agents, it limits how broadly the company can deploy them.\n\n## How it works\n\nRiften evaluates each request locally, applies the company’s cost, capability, latency, and reliability constraints, and selects among the models available to that organization.\n\nHard tasks can still reach frontier models. Routine work moves to lower-cost commercial or open-weight models. Teams can keep their provider accounts, pin a specific model when needed, and inspect the cost and routing decision behind every request.\n\nThe objective is not the cheapest token. It is the lowest expected cost of completing the work.\n\n## From routing to company-owned models\n\nThe router reveals which work is repeated. Outcomes reveal whether it was done correctly.\n\nA passing test, an accepted patch, a resolved ticket, or a completed workflow can become part of a customer-specific evaluation. Riften uses those evaluations to improve routing, post-train open models around recurring work, and test those models against the APIs they are intended to replace.\n\nModels earn production traffic through the same endpoint. Frontier APIs continue to handle novel work. Repeated workloads move onto models the company controls.\n\nThat creates a direct path from inference routing to evaluation, post-training, deployment, and the applications where the work happens.\n\nThe model labs are building intelligence every company can rent. Riften is building the path to intelligence every company can own.\n\n## Team\n\nRiften is founded by Carl Stauffer, a former Palantir Forward Deployed Engineer who shipped software into the purchasing and production systems of a nine-figure manufacturer. He was Palantir’s first Neurodivergent Fellow through Alex Karp’s program, conducted nuclear research at Jefferson Lab, studied computer science and applied mathematics at Johns Hopkins, and is an a16z Scout.\n\nThe rest of the team comes from Stanford Mathematics, Berkeley computer science, and Carnegie Mellon mathematics and computer science. Their work spans Oxford OATML, SPAR, CMU CyLab, Palo Alto Networks, Millennium, Amazon Search, NASA, UCSF Health, and industrial manufacturing systems.\n\nThey have built agent orchestration platforms, LLM evaluation and fine-tuning systems, cryptographic security software, production search infrastructure, CAD-matching systems, and autonomous QA for measuring agent performance. Their research includes an ICML acceptance and current work in review at NeurIPS.\n\nThe team also brings national-level recognition in mathematics and technical research spanning manufacturing, computational science, and new energy systems.\n\nRiften combines enterprise deployment, frontier evaluation research, competitive mathematics, agent systems, security, and production ML in one team built for the full path from request to outcome to model.\n\n## The ask\n\n**We are looking for companies with meaningful LLM usage across customer-facing products or internal agents.**\n\nIf most of your traffic still runs through one frontier model, send us:\n\n* Your current default model\n* Whether the traffic is internal or customer-facing\n* Your approximate monthly model spend or volume\n\nWe will show you what should remain on the frontier, what can move today, and whether Riften is worth testing.\n\nEmail us at [info@riften.ai](mailto:info@riften.ai).", "url": "https://www.ycombinator.com/launches/ShW-riften-route-every-ai-request-build-the-models-your-company-owns", "votes": 11, "created_at": "2026-08-17T15:12:03.010Z" } ] }, { "id": 32254, "name": "Prodigy Research", "slug": "prodigy-research", "former_names": [ "Prodigy AI" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/a2db1ee59cf597cd664444e784ae76e1f753d775.png", "website": "https://prodigy.markets", "all_locations": "San Francisco, CA, USA", "long_description": "Prodigy is the frontier AI trading research lab. We've trained the world's best foundation model for quantitative finance. We've already achieved more than 100% returns in live trading over the course of our YC batch, doubling our capital while major indices were flat. Our founders previously worked on a trading desk at Jane Street, trained frontier AI models at Google DeepMind and shipped cutting-edge AI at Apple to billions of customers. Join us as we use AI to solve the market.", "one_liner": "Training the world's best AI model for quantitative finance.", "team_size": 2, "industry": "Fintech", "subindustry": "Fintech", "launched_at": 1786153448, "tags": [ "Reinforcement Learning", "Finance", "Investing", "Trading", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "Fintech" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/prodigy-research", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/prodigy-research.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/prodigy-markets", "x_url": "https://x.com/realProdigyCorp", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/a2db1ee59cf597cd664444e784ae76e1f753d775.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Brad Flora", "url": "https://www.ycombinator.com/people/brad-flora" }, "founders": [ { "user_id": 3174989, "full_name": "Michael Wang", "title": "Founder/CEO", "founder_bio": "Founder and CEO at Prodigy Research (YC S26). Previously Senior Researcher at DeepMind, training and shipping AI models powering Google products with billions of monthly users. Before that, worked on a trading desk at Jane Street, one of the most prestigious and profitable trading firms in the world.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/michael-w-02810b161/", "x_url": "https://www.x.com/michaelwangelo", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/011ab6f302bb5ee939865ee4899d28ee9ea6acd8.jpg" }, { "user_id": 3203618, "full_name": "Yuhua Wang", "title": "Founder/CTO", "founder_bio": "Founder and CTO at Prodigy Research (YC S26). Previously shipped cutting-edge AI models and agents at Apple to billions of customers. Before that, built Salesforce's flagship AI platform serving hundreds of millions of user requests every day.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/yuhua-w-64348a66", "x_url": "https://x.com/yuhuawan", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/8ab0305694f8e67bd288110245604bf4db4e3a31.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 109747, "title": "Prodigy: The Frontier AI Trading Research Lab", "tagline": "Training the world's best foundation model for quantitative finance", "body": "**TL;DR** Prodigy Research is an AI lab that has trained the world's best foundation model for quantitative finance. We’ve already achieved 140% returns in live trading over the course of our YC batch, while major indices like the S&P 500 and the Nasdaq were flat or down.\n\n**Traction**\n\n![uploaded image](/media/?type=post&id=109747&key=user_uploads/3203618/c02ef51e-f0e4-43b8-a335-b55e66d5afd2)\n\nOur model beats Claude Fable and GPT-5.6 Sol at financial and quant research tasks. Prodigy's strategies are delta-neutral, and thanks to our rigorous and comprehensive risk controls, we've never had a down week.\n\n\n\n**Why Now**\n\nTrading firms generated hundreds of billions in profit last year, and this number is growing quickly YoY. Global financial markets now exceed $175 trillion in value across equities, derivatives, and digital assets. Prodigy's agents find new opportunities and alpha in the financial markets faster than any team of human quant researchers.\n\n![uploaded image](/media/?type=post&id=109747&key=user_uploads/3203618/31e0abc2-381b-4cdb-8d4f-86498b669b39)\n\n**The Team**\n\nOur founders are two brothers who left seven-figure jobs to build the future of quant trading. We bring together over 16 years of combined experience in quantitative finance, frontier model research, and large-scale AI infrastructure. Michael was a senior researcher at **Google DeepMind**, where he trained SoTA foundation models. He previously worked at **Jane Street**, one of the world’s most successful quantitative trading firms. Yuhua built cutting-edge AI at **Apple** serving **billions of users** across the world.\n\nA new era has arrived in the world of quantitative finance. To learn more, contact [founders@prodigy.markets](mailto:founders@prodigy.markets).", "url": "https://www.ycombinator.com/launches/SY7-prodigy-the-frontier-ai-trading-research-lab", "votes": 34, "created_at": "2026-08-10T18:30:00.360Z" } ] }, { "id": 32484, "name": "EdotEnv", "slug": "edotenv", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/a53586c9aacc3d5c0003f5bde0beb3535a3613f6.png", "website": "https://edotenv.com", "all_locations": "San Francisco, CA, USA", "long_description": "A Quant Neolab building toward RSI by leveraging markets as self-improving research environments.\r\n\r\nThey work with frontier AI labs and academic groups building research harnesses, evaluation benchmarks, and post-training environments.\r\n", "one_liner": "Quant Neolab building self-improving agents from quant data", "team_size": 2, "industry": "B2B", "subindustry": "B2B", "launched_at": 1786503507, "tags": [ "Reinforcement Learning", "Time Series" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/edotenv", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/edotenv.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/a53586c9aacc3d5c0003f5bde0beb3535a3613f6.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jared Friedman", "url": "https://www.ycombinator.com/people/jared-friedman" }, "founders": [ { "user_id": 2622651, "full_name": "Rui Wang", "title": "Founder/CEO", "founder_bio": "I am Rui (\"ray\" like the fish). CEO of EdotEnv, a quant neolab teaching LLMs robust research and good research taste. We use markets as an infinite source of increasingly hard data to build environments, harnesses, and post-train models.\nBackground in quant at G-Research, prev. applied maths at ETH Zurich.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/rui-wang-456032198/", "x_url": "https://x.com/RRicefan", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/c0681ade5300e908a4d3267d8ead62de1be2484e.jpg" }, { "user_id": 2995240, "full_name": "Michael Zhang", "title": "Founder/CTO", "founder_bio": "CTO at EdotEnv, a quant neolab building toward RSI.\n\nPreviously@Etched: Designed LLM Inference System for SOHU chips\nPreviously@TransmarketGroup: Built automated high-frequency option trading strategies\n\nETH Zürich Alumni", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/michael-zhang-845204232/", "x_url": "https://x.com/Michaelz1914", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/01fb39f6e87e8fc6884b4a54b36981b4011bb42a.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 110398, "title": "EdotEnv: Quant Neolab building towards RSI", "tagline": "We turn markets into self-improving research environments for AI agents.", "body": "EdotEnv (YC S26) is a Quant Neolab building toward RSI. They turn markets into self-improving research environments for AI agents.\n\nThe path to RSI is a continuous loop: agents conduct experiments, learn from the results, and autonomously become better at conducting the next round of research. For that loop to continue, each round must present a harder problem than the last.\n\nMarkets create this dynamic naturally. Every discovered inefficiency attracts competition, and every successful strategy makes future opportunities harder to find. Markets therefore provide an advancing frontier for self-improving agents.\n\nEdotEnv captures this process in multi-step environments where agents form hypotheses, design experiments, verify results and iterate. They use these environments to evaluate and post-train agents on increasingly difficult research tasks.\n\nThey work with frontier AI labs and academic groups building research harnesses, evaluation benchmarks, and post-training environments.\n\nEdotEnv's team previously worked in quantitative research at G-Research, high frequency options trading at TransMarket Group, and LLM inference research at Etched.\n\nIf you're working on RSI, RL, post-training, or AI for research, reach out to them at [info@edotenv.com](mailto:info@edotenv.com).\n\n![uploaded image](/media/?type=post&id=110398&key=user_uploads/89898/46b2eec2-b3c9-4467-b638-d9a51b2c39cd)\n\n", "url": "https://www.ycombinator.com/launches/Sic-edotenv-quant-neolab-building-towards-rsi", "votes": 5, "created_at": "2026-08-13T21:30:00.360Z" } ] }, { "id": 32554, "name": "Ooak Data", "slug": "ooak-data", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/ad1b7cce12014a7746a7f8b912560717f6e2f037.png", "website": "https://ooakdata.com/", "all_locations": "San Francisco, CA, USA", "long_description": "Ooak Data turns internal company data into a training ground where AI agents can be tested on real work.\r\n", "one_liner": "We turn company data into training data", "team_size": 5, "industry": "B2B", "subindustry": "B2B", "launched_at": 1783922301, "tags": [ "Reinforcement Learning", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": true, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada", "Remote", "Partly Remote" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/ooak-data", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/ooak-data.json", "year_founded": 2024, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/107463604", "x_url": "https://x.com/Ooakdata", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/ad1b7cce12014a7746a7f8b912560717f6e2f037.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Diana Hu", "url": "https://www.ycombinator.com/people/diana-hu" }, "founders": [ { "user_id": 3207024, "full_name": "Pierre-Louis Vouteau", "title": "Founder/CEO", "founder_bio": "Cofounder & CEO at Ooak Data (YC S26). \nPrior COO @Visely (UK ed-tech)\nex Gopuff and L'Oréal", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/pierre-louis-vouteauau-15a06566/", "x_url": "https://x.com/PLouisV", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/465fc31f2866a593a25e59c4839739aafddbb167.jpg" }, { "user_id": 3207121, "full_name": "Thomas Aubry", "title": "Founder/CTO", "founder_bio": "Founder CTO at OoakData. \nPreviously Head of Data at PayLead, where he led the Data & AI team. Applied ML Scientist at Samsung AI. Earlier co-founded Macro Vision on autonomous vehicles Engineering. \nEducated at Institut supérieur de mécanique de Paris in Mechatronics and Robotics.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/thomaub", "x_url": "https://x.com/ThomAub", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/1cb368f1fd2593cea6dfaa5e585238e725907fcc.jpg" } ], "jobs": [ { "id": 96721, "title": "ML engineer", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/M6anRxc-ml-engineer", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "€70K - €100K EUR", "equity_range": "0.10% - 0.30%", "experience": "6+ years", "visa": "US citizenship/visa not required", "skills": [ "Machine Learning", "Reinforcement learning (RL)" ] }, { "id": 99220, "title": "Software Engineer Senior ", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/KqZT9J1-software-engineer-senior", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "€55K - €80K EUR", "equity_range": "0.10% - 0.30%", "experience": "6+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 99221, "title": "Data engineer senior", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/0CZAefT-data-engineer-senior", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Engineering", "role_type": "Backend", "salary_range": "€65K - €95K EUR", "equity_range": "0.10% - 0.30%", "experience": "3+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 99325, "title": "Founding Marketing Lead ", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/atIhXqo-founding-marketing-lead", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Marketing", "role_type": null, "salary_range": "€50K - €80K EUR", "equity_range": "0.10% - 0.50%", "experience": "6+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 107967, "title": "1st Product Manager ", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/GqbqtyS-1st-product-manager", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Product", "role_type": null, "salary_range": "€70K - €90K EUR", "equity_range": "0.10% - 0.30%", "experience": "6+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 112051, "title": "Head of Engineering ", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/cFitSes-head-of-engineering", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Engineering", "role_type": "Full stack", "salary_range": "€100K - €130K EUR", "equity_range": "", "experience": "11+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 114401, "title": "Account Executive - Data Partnerships (US)", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/q01Cgyi-account-executive-data-partnerships-us", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$100K - $140K", "equity_range": "", "experience": "6+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 114402, "title": "GTM Manager (US)", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/BJ8W0DT-gtm-manager-us", "location": "San Francisco, CA, US", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "$120K - $150K", "equity_range": "", "experience": "6+ years", "visa": "US citizen/visa only", "skills": [] }, { "id": 114403, "title": "Head of Operations", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/fl4Oyho-head-of-operations", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "€80K - €100K EUR", "equity_range": "", "experience": "11+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 114404, "title": "Operations Manager - Data & AI", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/j3mCCMu-operations-manager-data-ai", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "€45K - €55K EUR", "equity_range": "", "experience": "1+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 114405, "title": "Data Quality Specialist", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/Gr7Gc1x-data-quality-specialist", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Operations", "role_type": null, "salary_range": "€65K - €80K EUR", "equity_range": "", "experience": "6+ years", "visa": "US citizenship/visa not required", "skills": [] }, { "id": 116458, "title": "Data Partnerships - Account Executive (EU)", "url": "https://www.ycombinator.com/companies/ooak-data/jobs/S7N4ejR-data-partnerships-account-executive-eu", "location": "Paris, IDF, FR / Paris, Île-de-France, FR", "type": "Full-time", "role": "Sales", "role_type": null, "salary_range": "€55K - €70K EUR", "equity_range": "", "experience": "3+ years", "visa": "US citizenship/visa not required", "skills": [] } ], "news": [], "launches": [ { "id": 110403, "title": "Ooak Data: Turn internal company data into a training ground where AI agents can be trained on real work", "tagline": "The bottleneck to smarter agents isn't compute. It's real data. We unlock it.", "body": "**TL;DR** Ooak Data turns internal company data into a training ground where AI agents can be tested on real work.\\\n\\\n[https://drive.google.com/file/d/1EgbFzN-g4cVY-dk7ZPH6gR37F28D56xo](https://drive.google.com/file/d/1EgbFzN-g4cVY-dk7ZPH6gR37F28D56xo/view?usp=sharing)\n\n**The problem**\n\nAI agents still fail on real work: the messy, multi-tool, permission-heavy workflows inside actual companies. You can't synthesize it either, real org charts, permissions, and cross-tool dependencies don't come out of a generator. You have to source it from real companies and make it safe. That's the hard part, and it's what we do.\n\n**Our Solution**\n\nWe connect to a company's tools (Gmail, Slack, Notion, Jira, Drive, SharePoint) and anonymize everything into a digital twin: structurally identical, but privacy-safe. Labs use our product to train and evaluate AI agents on real work.\n\nWe're working with multiple frontier labs and are now well above 7 figures in revenue. We've cut ingestion-to-delivery time 3x since launch, across data from 20 companies.\n\n**The team**\n\nWe’re a team of 3 cofounders: Pierre-Louis, Grégoire and Thomas and spent our careers inside the messy company systems we now replicate.\n\n**Our asks**\n\n* If you're training or evaluating agents on real workflows, we'd love to talk.\n* Connections with companies abo 20+ FTEs, operating, winding down, or recently closed. Know someone who fits? **Add them [here](https://form.typeform.com/to/PjHdZmDW)**, or if it's your own company, grab [**15 min**](https://calendar.google.com/appointments/schedules/AcZssZ142jxQGEjB1PR9skAe17Nf7Pjf8G8EszSPcRhZIzmZXGCkn05REmmAmqPvMgWTYUMwbQNwinIF). Curious what a dataset is worth? Try our [**simulator**](https://ooakdata.com/partner).\\\n feel free to reach out to us: [pierrelouis@ooakdata.com](mailto:pierrelouis@ooakdata.com)\n\n![uploaded image](/media/?type=post&id=110403&key=user_uploads/3207024/bddd7844-32fc-4746-a3bd-e2a8450a85bd)\n\nYou can also reach us here: [pierrelouis@ooakdata.com](mailto:pierrelouis@ooakdata.com)", "url": "https://www.ycombinator.com/launches/Sih-ooak-data-turn-internal-company-data-into-a-training-ground-where-ai-agents-can-be-trained-on-real-work", "votes": 26, "created_at": "2026-08-14T15:00:00.284Z" } ] }, { "id": 32624, "name": "Fabraix", "slug": "fabraix", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/d3d83aa2013bbe6b22b37f316b294d0c717eea35.png", "website": "https://fabraix.com", "all_locations": "San Francisco, CA, USA", "long_description": "Fabraix builds state-of-the-art AI red-teaming agents that continuously detect security vulnerabilities in customer-facing AI. Our product, Nyx, has already found vulnerabilities in agents at dozens of Fortune 500 companies. On AgentHarm, the leading benchmark for offensive AI security, Nyx achieved a 78% attack success rate, compared with 67% for GPT-5.6 Sol.\r\n\r\nAI agents can change more often than teams can test them. A new model, prompt, tool, permission, or data source can change what an agent does, even when the application code stays the same. And AI is also increasing how much software companies produce and how often it changes. We built Nyx to automate the work required to red-team AI agents. It is multi modal by design and interacts with the target indirectly through controlled replicas of SaaS products and websites that we maintain, placing malicious payloads in webpages, documents, files, messages, and tool outputs to test how an agent behaves under different conditions.\r\n\r\nNyx often finds its first vulnerability within minutes or hours rather than days or weeks. Because the work is fully automated, companies can repeat the test with every change and at a much lower cost than what a 7-figure red-teaming (and point-in-time) engagement would require.\r\n\r\nWe're also the team behind ACE (Adversarial Cost to Exploit), a benchmark that measures AI security in terms of how much it costs attackers to break an AI system; providing a game-theoretic framework to understand how motivated a rational attacker would be in exploiting the system.", "one_liner": "The world's frontier hacker of AI agents.", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Security", "launched_at": 1783913331, "tags": [ "Reinforcement Learning", "Cybersecurity", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Security" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/fabraix", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/fabraix.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/fabraix/", "x_url": "https://x.com/FabraixHQ", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/d3d83aa2013bbe6b22b37f316b294d0c717eea35.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Jon Xu", "url": "https://www.ycombinator.com/people/jon-xu" }, "founders": [ { "user_id": 693526, "full_name": "Ahmed Aly", "title": "Founder/CEO", "founder_bio": "Co-founder @ Fabraix, building a frontier hacker agent that finds failure modes and security exploits in AI systems before real users do. Previously the first data scientist at Sequoia-backed Two, where he built AI systems for adversarial environments from scratch and scaled them to process $1B+ in annual B2B transactions, averting $50M in losses for multiple Fortune 150 companies. A published researcher with multiple papers in Q1 journals and a UCL PhD drop-out.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ahmed-a-657264b4/", "x_url": "https://x.com/zachdotai", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/72bc6e2cde8727671fbf9c026c0c9a0ffcc80f4f.jpg" }, { "user_id": 2769430, "full_name": "Ibrahim Abdu", "title": "Founder/CTO", "founder_bio": "Red-teamed AI agents in Meta's Privacy org; built an SRE agent triaging 3M+ errors/month, saving 10,000+ engineering hours a year. Built a compiler for insider-trading detection at TradingHub: $100M+ in daily trades; Oxford PPE, top 8% of cohort.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/i-a-7352b7137/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e3d923fd8bbb4ccfb176340c73f6ebc968516321.jpg" } ], "jobs": [], "news": [ { "title": "Meet The YC Startups Betting On What Comes Next", "url": "https://www.forbes.com/sites/dariashunina/2026/09/03/meet-the-yc-startups-betting-on-what-comes-next/", "date": "Sep 03, 2026" } ], "launches": [ { "id": 106270, "title": "Fabraix: The Frontier Agent that Hacks Customer-facing AI.", "tagline": "Fabraix builds AI red-teaming agents that continuously detect security vulnerabilities in customer-facing AI. We already found vulnerabilities in agents at dozens of Fortune 500 companies.", "body": "**TL;DR**: Fabraix builds AI red-teaming agents that continuously detect security vulnerabilities in customer-facing AI. Our product, Nyx, has already found vulnerabilities in agents at dozens of Fortune 500 companies, and hits a 78% attack success rate on AgentHarm (leading offensive AI security benchmark) vs. 67% for GPT-5.6 Sol.\n\n**Ask**: If your team ships customer-facing AI agents (chat, voice, browser, or coding) and needs to continuously test them for security vulnerabilities, find us at [founders@fabraix.com](mailto:founders@fabraix.com).\n\n\n\n**The Problem**\n\nAI agents introduce new vulnerabilities not present in traditional software..\n\nIn order for agents to be truly useful, they must be trusted to interact with untrusted surfaces like MCPs, websites and files, take real actions, and have access to confidential data. You can’t tell your defenses are sound by just reading your prompts since the agent's behaviour is not deterministic. The best thing to do is run your agent and really try to break it.  \n\nDoing this red-teaming process manually is time consuming, costly, and you are unlikely to catch everything. And as soon as the agent is updated, you need to repeat it. Security teams simply don’t have the bandwidth to keep up with the rapid deployment inside organisations.\n\n**Our Solution**\n\nNyx is our red-teaming agent that tests customer-facing AI agents, via direct interaction and also indirectly via its environment. It requires no access to your agent’s source code in order to test.\n\nNyx is highly adaptive when attacking. Rather than cycling through hard coded payloads like existing tools, it dynamically adjusts its attack strategies based on your agents defenses and responses. It probes your agent over a wide surface area of attack vectors, and deep dives This adaptive approach makes it 20x more effective as a red-teamer compared to other solutions.\n\nWe maintain a library of over 10,000 attack strategies and jailbreaks. These are high level strategies which are used by Nyx to generate dynamic adaptive attacks, over multiple turns. \n\nEvery finding includes the attack steps, the agent’s responses, and the resulting failure for you to reproduce. You can then rerun the same test after a release to check whether the failure remains fixed. \n\nOn the AgentHarm benchmark, Nyx achieved a 78% attack success rate, compared with 67% for GPT-5.6 Sol. Nyx has also found exploitable failures in public-facing agents operated by dozens of Fortune 500 companies.\n\nWe also built ACE (Adversarial Cost to Exploit), a benchmark that measures AI security in terms of how much it costs attackers to break an AI system; providing a game-theoretic framework to understand how motivated a rational attacker would be in exploiting the system.\n\n**The Team**\n\nWe both have a lot of experience building and securing AI agents at scale.\n\nIbrahim built AI agents at Meta that diagnosed and fixed production errors. Before that, he built compilers and database engines in Fintech. He graduated from Oxford in the top 8% of his cohort.\n\nAhmed led international payments fraud team at Monzo. Before that, he was the first data scientist at a Sequoia-backed startup, where he built its fraud engine from scratch into processing more than $1B in annual B2B transactions and preventing tens of millions in losses for Fortune 150 companies. He has published several peer-reviewed research papers in Q1 journals.\n\nWe previously worked together, and have known each other for almost four years.\n\n![uploaded image](/media/?type=post&id=106270&key=user_uploads/693526/5f437cbd-2009-40cc-8c69-746da9ea8300)\n\n**The Ask**\n\nIf you ship a customer-facing (chat, voice, browser, or coding) agent, we can help you find all the failure modes and built a self-healing loop to discover, triage, and maintain secure AI agents.\n\n[founders@fabraix.com](mailto:founders@fabraix.com)", "url": "https://www.ycombinator.com/launches/Re2-fabraix-the-frontier-agent-that-hacks-customer-facing-ai", "votes": 127, "created_at": "2026-07-20T15:06:58.917Z" } ] }, { "id": 32988, "name": "Markov", "slug": "markov", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/41a4a41e5482735389d2089a01ff2b862f56e307.png", "website": "https://www.markovstudios.com/", "all_locations": "San Francisco, CA, USA", "long_description": "Markov builds expert computer-use datasets for frontier AI labs. Our datasets include screen recordings + mouse/keyboard actions of professionals using CAD, design and enterprise software.\r\n\r\nWe work with the top AI labs and we’ve gotten more than 350k downloads on HuggingFace. Our long-term goal is to be the full stack platform for computer-use AI, starting with data and then the infrastructure to train and deploy computer-use models.\r\n\r\nDev and Harish are technical co-founders. Dev studied aerospace engineering at IIT Madras and was the youngest team member at Sarvam - India’s top AI lab. Harish is an Emergent Ventures grantee and his prior robotics work was featured in top national publications.", "one_liner": "Expert computer-use data for frontier AI labs", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1782277833, "tags": [ "Reinforcement Learning", "Data Labeling", "Data Engineering" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/markov", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/markov.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/markov-artificial-intelligence/", "x_url": "https://x.com/markov_labs", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/41a4a41e5482735389d2089a01ff2b862f56e307.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Harshita Arora", "url": "https://www.ycombinator.com/people/harshita-arora" }, "founders": [ { "user_id": 1783684, "full_name": "Dev Mandal", "title": "Founder/CEO", "founder_bio": "Co-Founder & CEO at Markov. Prev at Sarvam, Aerospace engineering at IIT Madras", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/dev-mandal-iitm/", "x_url": "https://x.com/DevvMandal", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/5a53c70cbcddfae2c6c21e32faac5865be5702b2.jpg" }, { "user_id": 2921049, "full_name": "Harish Ashok", "title": "Founder/CTO", "founder_bio": "Co-Founder at Markov. Previously founded Zenith, a robotics company building AI-assisted hardware workflow tools. ", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/harish-ashok", "x_url": "https://x.com/harishashok", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/fbfaea52b2a5bc846c87d04392062585acfb69ff.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 109874, "title": "Markov - Data for computer-use AI", "tagline": "150K+ downloads across our HuggingFace datasets", "body": "# \\*\\*\\*\\*\n\n**TLDR;**\n\n* **Markov builds the data infrastructure for computer-use AI**, collecting synchronized screen recordings, mouse/keyboard actions, and annotations from real workflows.\n* **We specialize in hard-to-source expert data** **across CAD, architecture, engineering**, and other professional applications.\n* **We’ve already sold 15,000+ hours of data to frontier AI labs** and open-sourced leading computer-use datasets with **150k+ downloads on HuggingFace**.\n\n\n\n**The Problem:**.\n\nWhile frontier models have gotten much better at reasoning, they still struggle to reliably complete long, multi-step workflows across real desktop software. **The biggest bottleneck is the lack of high-quality training data.**\n\nTraining these models requires **synchronized screen recordings, mouse and keyboard actions, and high-quality annotations from people doing real work**. This is especially scarce for expert workflows in domains like CAD, architecture, engineering, finance, and other professional software.\n\n**That’s where we come in:**\n\nWe work with a vetted network of expert contributors to collect high-quality computer-use data across:\n\n* Consumer applications\n* Productivity software\n* Accounting and finance\n* CAD, architecture, engineering, and other expert workflows\n\n![uploaded image](/media/?type=post&id=109874&key=user_uploads/1783684/ba673ff4-e489-4753-b11b-4797762834fb)\n\n**So far we've:**\n\n* Sold more than 15,000 hours of data to frontier AI labs\n* Open-sourced the [**largest dataset of computer-use screen recordings**](https://x.com/DevvMandal/status/2032104265121140924?s=20)(**150k+ downloads on HuggingFace**)\n* Released the [**largest open gaming computer-use dataset**](https://x.com/DevvMandal/status/2072357115847671990?s=20)\n\n**Our Ask:**\n\nIf you're building computer-use agents and need high-quality training data - or have interesting workflows you'd like collected - we'd love to chat **:)**\n\nBook a meeting with us at Or shoot us an email at [founders@markovstudios.com](mailto:founders@markovstudios.com)", "url": "https://www.ycombinator.com/launches/SaA-markov-data-for-computer-use-ai", "votes": 11, "created_at": "2026-08-11T18:00:00.277Z" } ] }, { "id": 33022, "name": "Praxis Robotics", "slug": "praxis-robotics", "former_names": [ "Praxis AI" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/9bc18b804ad9a5d92ef1db88e90491f98e0204a7.png", "website": "https://www.praxisrobotics.io", "all_locations": "San Francisco, CA, USA", "long_description": "Praxis captures the data physical AI is bottlenecked on: egocentric video, 3D scans, and multimodal capture from inside real industrial and residential environments. Embedded within publicly listed and unicorn-scale conglomerates, we reach 60k workers across 5 continents and 150+ environment types, supplying the real-world training data frontier labs and humanoid companies can't source anywhere else.", "one_liner": "Turns real businesses into robotics data vendors", "team_size": 3, "industry": "Industrials", "subindustry": "Industrials", "launched_at": 1782948528, "tags": [ "Hard Tech", "Reinforcement Learning", "Robotic Process Automation" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "Industrials" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/praxis-robotics", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/praxis-robotics.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/130274179/admin/dashboard/", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/9bc18b804ad9a5d92ef1db88e90491f98e0204a7.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Diana Hu", "url": "https://www.ycombinator.com/people/diana-hu" }, "founders": [ { "user_id": 2794157, "full_name": "Rohan Seelamsetty", "title": "Founder/CEO", "founder_bio": " CEO @ Praxis. Forged across factory floors, campaign trails, and niche forums. Ex Computation and Cognition at Penn, Apollo Fellow at Oxford, and Indian Debate Team (#1 ranked team worldwide)", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/rohan-s-1186b2214", "x_url": "https://x.com/ro_seelamsetty", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/4b3e6e16cddb3da2384ce8f9439ca9ebc043d2bb.jpg" }, { "user_id": 3752185, "full_name": "Dev Karpe", "title": "Founder/COO", "founder_bio": "Founder at Praxis AI, previously at Wharton's Huntsman Program. Background in investment banking and supply chain ops at a F500 CEO's office. ", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/dev-karpe", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/f4a1fc8056a887a99633492428b5db995e58a488.jpg" }, { "user_id": 3748464, "full_name": "Tommy Li", "title": "Founder/CTO", "founder_bio": "Founder at Praxis AI. Prev. tech private equity at Morgan Stanley. Industrial tech venture capital at Andreessen Horowitz. Published IEEE researcher at NASA JPL and PerceptIn Robotics. BSE CS and Finance from UPenn.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/tommyli-penn/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/0e7be9454dacfd8fa65594cc6836b90cc7e5e0b4.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 109471, "title": "Praxis AI - Training Data for Robotics", "tagline": "Praxis offers labs the most diverse environments, SOTA hardware, and rigorous post processing to train their models", "body": "**Hey YC!**\n\nWe're Rohan, Dev, and Tommy, and we're building Praxis.\n\n**Praxis turns real-world work into training data for robots.**\n\n\n\n**What we do**\n\nRobots learn to act in the physical world by watching humans do the work. We partner with businesses to capture egocentric human demonstration data, the first-person video of how tasks actually get done in factories, warehouses, homes, and other physical work environments, then supply it to the labs and humanoid teams training the next generation of robots. Robotics data is our focus, but since our collection infrastructure overlaps, we're happy to explore other multimodal data too.\n\n**The problem**\n\nRobotics is missing its internet-scale training corpus. LLMs learned from the web, books, code, and the accumulated record of human thought. Robots need the equivalent record of human action: how people pick, pack, clean, assemble, inspect, cook, use tools, recover from mistakes, and move through messy physical environments.\n\nCollecting some of this data is easy. Collecting enough of the right kind is brutally hard, for two reasons.\n\nFirst, diversity is capped by access. Models only generalise if they've seen the full range of real environments and operators, from a one-person shop to a national-scale industrial operation. Getting in the door across that entire spectrum, and across continents, is a physical-world access problem that most teams can't solve.\n\nSecond, it takes two rare skills at once. You have to know what frontier labs need at a technical level, and be able to run demanding physical operations across thousands of real sites, standardising capture and turning fragmented workflows into model-ready data. Few teams have both halves.\n\n**Our solution**\n\nWe already run collections across 150 environment types on 5 continents, with diversity from mom-and-pop stores to conglomerates accounting for 3% of a nation's GDP, and a reachable network of 60,000 workers. Depending on the task, we capture egocentric video, stereo-RGB, IMU, wrist-camera views, glove kinematics, haptics, and dense annotations. Everything is delivered calibrated, synced, pose-tracked, QA'd, and PII-scrubbed. We build and deploy custom capture hardware quickly because part of the team operates out of Shenzhen.\n\n**Why us**\n\nPraxis is built to do both hard things at once. We pair robotics research depth (IEEE, NASA JPL) with the ability to actually run demanding operations in the real world, from Fortune 500 supply chains to Parliaments, plus a Shenzhen hardware team. That combination- technical understanding of what labs need plus real physical-world access at scale- is what lets us deliver diverse, model-ready data at a cost and breadth that is irreplaceable.\n\n**Our asks**\n\n1. Intros to anyone working on robotics, embodied AI, or world-models, especially data and research leads. If you need other forms of multimodal data, we're happy to explore that too.\n2. If you run a business with interesting physical work (logistics, manufacturing, food, retail, or lab work) and would like to monetize it, we'd love to talk.\n\nYou can reach us at [founders@praxisrobotics.io](mailto:founders@praxisrobotics.io) and [rohan@praxisrobotics.io](mailto:rohan@praxisrobotics.io)[ https://www.praxisrobotics.io](https://www.praxisrobotics.io)", "url": "https://www.ycombinator.com/launches/STf-praxis-ai-training-data-for-robotics", "votes": 11, "created_at": "2026-08-06T22:30:00.618Z" } ] }, { "id": 33051, "name": "Tokenless", "slug": "tokenless", "former_names": [ "Tokenless", "Touchy" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/7f250ec6cc81104148d8bd5bafdd6b96bc77d023.png", "website": "https://usetokenless.com", "all_locations": "San Francisco, CA, USA", "long_description": "Tokenless is an applied AI research lab. Our goal is to end reward hacking. We do research and build tools to detect issues with environments and evals, understand model behavior in those environments, and prevent those issues from happening. Correcting them makes training more effective, makes model performance easier to interpret, and is crucial to solving the AI alignment problem.\r\n\r\nIf you’re training models with RL, building environments, or running agent evals, and want more confidence in your environments, we're here to help.", "one_liner": "We find flaws in RL environments and evals and help you fix them.", "team_size": 3, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1786732497, "tags": [ "Developer Tools", "Machine Learning", "Reinforcement Learning", "B2B", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/tokenless", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/tokenless.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/usetokenless/", "x_url": "https://x.com/usetokenless", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/7f250ec6cc81104148d8bd5bafdd6b96bc77d023.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/579511aa3dc1cfb92c5f713907f31dd6fe36c640.png" ], "primary_partner": { "name": "Vivian Midha Shen", "url": "https://www.ycombinator.com/people/vivian-midha%20shen" }, "founders": [ { "user_id": 3697324, "full_name": "Rohit Agarwal", "title": "Founder", "founder_bio": "Founder at Tokenless (YC S26). CS PhD student at Princeton PLI, Hazan Lab, Google Deepmind. Previously at Duolingo and theory research at Berkeley, was EECS #1 BS grad in 2024.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/rohit-agarwal-b647951ba/", "x_url": "https://x.com/Rohit_Writes", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/86b74b1209c81d1d158af5587de149a3e8c2292c.jpg" }, { "user_id": 3706728, "full_name": "Andrew Liu", "title": "Founder", "founder_bio": "Founder at Tokenless (YC S26). Formerly a researcher at UC Berkeley working on language documentation tools. BA in Computer Science, Mathematics, and Linguistics from UC Berkeley. Previously TA'd CS61C.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/andrew-liu-b50873261/", "x_url": "https://x.com/landrew7363", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/04017660afe1efc138b242cbcc7c4efffc5f5bc4.jpg" }, { "user_id": 3732665, "full_name": "Kevin Wang", "title": "Founder", "founder_bio": "Founder at Tokenless. Prev MTS at SkyPilot building multi-cloud GPU infra for top neolabs, Founding Engineer at Eventual from seed to Series A building the data engine for physical AI, EECS at UC Berkeley.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/kevinzhiwang/", "x_url": "https://x.com/kevzwang", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/1a21a2cf6802cfdcf6769408ee9d25a254858622.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 111407, "title": "Touchy 🙌: Voice-first AI Assistant for iPhone that understands the world around you", "tagline": "Your personal AI assistant that sees what you see, remembers what matters to you, and gets things done across all your tools. Just ask out loud.", "body": "**TL;DR:** We’re building Touchy 🙌, a voice and image-first AI assistant for the iPhone that personalizes itself to you. You can activate Touchy from your lock screen to instantly get things done across your apps or get answers about the world around you. Touchy is free to try, get it here: \n\n\n\n## **The Problem**\n\nYou probably know the feeling:\n\nYou ask the assistant on your phone, across the room, to send a colleague a message.\n\nIt cheerfully replies that it found 10 search results for “text Bob on Slack”.\n\nSo you trudge to your phone, find the app, open the DM, and begrudgingly type out your message instead.\n\nWe wanted a better personal assistant, one that had the intelligence and integrations we’ve come to expect from our AI agents, with the convenience of one that was native to your devices. We also believe that for AI to truly be personal, it needs to be deeply integrated into your life and the world around you. Right now, AI systems are constrained largely to using digital tools, yet the physical world is rich with audio and visual information that your phone is capable of receiving.\n\n## What We Built\n\nSo we built Touchy. One button press and you're instantly talking to an assistant that can see what you see, remember what you've said, and take action across all the tools you use. \n\nWhat sets Touchy apart:\n\n* **Reimagined UX.** Touchy is a fast, fully iOS-native app with a unique interface that is delightful to use and brings it closer to the world around you.\n* **Sees and hears the world around you**. Speak to Touchy and it can speak back. Easily snap a photo with the app’s camera to give Touchy access to what you’re seeing.\n* **Gets things done across your apps**. Touchy comes with built-in integrations with 1,000+ popular apps, as well as iOS-native capabilities such as you contacts and location.\n* **Make it yours over time**. You chat with Touchy in a single, continuous conversation. It automatically collects memories and preferences, and tailors itself to be more useful to you as you use it.\n\nIt's built to give you your attention back. Ask it something, let it act, and get back to your day — no app fighting for your attention.\n\n![uploaded image](/media/?type=post&id=111407&key=user_uploads/3697324/23ce7916-fce5-44c4-9ebf-f04049fee639)\n\n## The team\n\nWe’re Rohit, Andrew, and Kev. Before this, Rohit was doing a PhD on AI alignment at Google DeepMind Princeton and graduated #1 in Berkeley EECS. Andrew built educational software that scaled to half a million students worldwide. Kev was a startup founding engineer, developing data and GPU infrastructure that power many of the leading AI neolabs today. \n\nWe met during undergrad at UC Berkeley, and we’re coming back together to research, productionize, and scale the future of personal technology.\n\n## Our ask\n\nIf you have an iPhone, download Touchy and try it out: \n\nWe would love to learn what people find themselves using Touchy for, and what they wish they could use it for. You can also report bugs by shaking your phone or just telling Touchy. You can also reach us at [founders@touchyapp.com](mailto:founders@touchyapp.com)", "url": "https://www.ycombinator.com/launches/Syt-touchy-voice-first-ai-assistant-for-iphone-that-understands-the-world-around-you", "votes": 24, "created_at": "2026-08-20T21:00:00.530Z" } ] }, { "id": 33310, "name": "Petrarch", "slug": "petrarch", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/091a0f9893d5a0f66b65399660fc8041716acd4e.png", "website": "https://petrarch.co/", "all_locations": "San Francisco, CA, USA", "long_description": "Petrarch brings internal company data to frontier labs, starting with manufacturing and industrial companies. We source data including codebases, project files, and payments from bankruptcy courts, then de-identify and prepare the data for model training. \r\n", "one_liner": "Internal industrial company data for frontier labs", "team_size": 3, "industry": "B2B", "subindustry": "B2B", "launched_at": 1782965522, "tags": [ "Reinforcement Learning", "B2B", "Privacy", "Data Engineering", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/petrarch", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/petrarch.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/petrarchinc", "x_url": "https://x.com/petrarchinc?s=11", "fb_url": null, "cb_url": "https://www.crunchbase.com/organization/petrarch-e3b1", "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/091a0f9893d5a0f66b65399660fc8041716acd4e.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Vivian Midha Shen", "url": "https://www.ycombinator.com/people/vivian-midha%20shen" }, "founders": [ { "user_id": 3655939, "full_name": "Ian Lee", "title": "Founder/CEO", "founder_bio": "Founder at Petrarch. Previously Fellow at Meridian AI building training data and GTM infrastructure for AI-native Excel in financial services. Technical Advisor at Scale AI, working on quality filtering and annotation for agentic assistants and reasoning models. BA in Applied Mathematics and Psychology from Harvard.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/-ianlee", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/bceb6907c6226831472da7b1d7688785c1fb3c17.jpg" }, { "user_id": 2670953, "full_name": "Samuel Hahn", "title": "Founder/COO", "founder_bio": "Founder at Petrarch. Previously led GTM at Magier AI (Techstars '24), building agentic PII-redaction and data-compliance products. Venture analyst at Mainstreet sourcing pre-seed fintech and deeptech. BA in Economics and History from Harvard.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/samuelhahn9/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/c4c5f0acb76bd4a03ee95588c1e5324406ca2dae.jpg" }, { "user_id": 3594340, "full_name": "Sudhish Swain", "title": "Founder/CTO", "founder_bio": "Founder at Petrarch. Previously built computer vision software for medical equipment identification deployed in California hospitals as Head of Software Engineering at Clean Sweep Group. Research intern at Brigham and Women's Hospital analyzing spatial transcriptomic cancer datasets. BA Integrative Biology at Harvard.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/sudhish-swain/", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/098958191b28eaf5108b38cb0d2768964c348d55.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 105007, "title": "Petrarch — Unlocking enterprise data for tomorrow's economy", "tagline": "The market for specialized data locked up in enterprise companies.", "body": "# The market for specialized data locked up in enterprise companies.\n\n[Petrarch](http://petrarch.co) is unlocking enterprise data for tomorrow's economy.\n\n\n\nFrontier labs scraped the internet dry. Then, labelers contracted subject matter experts to hand-make data. Now demand has shifted towards the bespoke data locked up within industries like construction, manufacturing, and logistics.\n\nPetrarch targets companies with valuable **operational** **data** they haven't monetized, from **legacy** **businesses** and **startups** to companies undergoing **bankruptcy** and **pivots**.\n\nPetrarch **sources** suppliers, **cleans** and **restructures** the datasets, certifies **legal clearance**, and resells to **applied AI startups and labs.** They have already transacted construction plans, autonomous driving recordings, and codebases.\n\nWith experience at Scale and Meridian AI, founders Ian Lee, Samuel Hahn, and Sudhish Swain dropped out of Harvard to unlock the next wave of training data.\n\nIf you need proprietary datasets or think you have something valuable on your hands \\\n(or recently pivoted), reach out at [founders@petrarch.co](mailto:founders@petrarch.co) or visit us at [Petrarch.co](http://Petrarch.co)!", "url": "https://www.ycombinator.com/launches/RJf-petrarch-unlocking-enterprise-data-for-tomorrow-s-economy", "votes": 11, "created_at": "2026-07-13T16:11:02.837Z" } ] }, { "id": 33345, "name": "CueBench", "slug": "cuebench", "former_names": [ "CueBench", "Snap Poker" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/8fc25d9ffc54b20b2ccee87e57733b12a6e8edf1.png", "website": "https://cuebench.dev", "all_locations": "San Francisco, CA, USA", "long_description": "", "one_liner": "RL post-training for science reasoning and performance engineering", "team_size": 3, "industry": "B2B", "subindustry": "B2B", "launched_at": 1786744876, "tags": [ "Reinforcement Learning", "Data Engineering" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/cuebench", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/cuebench.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/8fc25d9ffc54b20b2ccee87e57733b12a6e8edf1.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Vivian Midha Shen", "url": "https://www.ycombinator.com/people/vivian-midha%20shen" }, "founders": [ { "user_id": 3732725, "full_name": "Dillon Mehta", "title": "Founder/CEO", "founder_bio": "Founder and CEO at CueBench", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/dillon-mehta-49aa2827b/", "x_url": "https://x.com/PickleDilll_", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/812fc022db435026c7bee56ba58f916bfd5218ee.jpg" }, { "user_id": 3762638, "full_name": "Neel Gadde", "title": "Founder/COO", "founder_bio": "Co-Founder and CRO at CueBench", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/neel-gadde-491880377/", "x_url": "https://x.com/neelapolitan?s=11", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/d2be7502db72b4faa82985dd5181743eabe8a5aa.jpg" }, { "user_id": 3791688, "full_name": "Rishan Hemrajani", "title": "Founder", "founder_bio": "Co-Founder & CTO of CueBench.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/rishan-hemrajani-254bba292", "x_url": "", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e3bd268e04f9551b93d77346c5652657d25a2b69.jpg" } ], "jobs": [], "news": [], "launches": [] }, { "id": 33348, "name": "Autoresearch, Inc.", "slug": "autoresearch-inc", "former_names": [ "hiloop" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/5b91efea37962386a163af58e2f2920b1e9d0aa7.png", "website": "https://autoresearch.inc", "all_locations": "San Francisco, CA, USA", "long_description": "Autoresearch helps teams train agents for tasks where general models are not good enough. Give us a task, your current agent or model, and an evaluation. Autoresearch runs an autoresearch campaign across data, SFT and other post-training methods, continual learning, prompts, tools, harnesses, and systems, then returns the best verified improvement. It runs hosted or in your cloud.\r\n\r\nWe provide the research system around models: persistent memory, full experiment lineage, compute orchestration, and statistical verification. We’re starting with agent and model training, continual learning, and optimization.\r\n\r\nReach out to us for early access at founders@autoresearch.inc.", "one_liner": "Autoresearch as a service: we solve your hardest measurable problems", "team_size": 3, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1785348094, "tags": [ "Developer Tools", "Machine Learning", "Reinforcement Learning", "Infrastructure", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/autoresearch-inc", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/autoresearch-inc.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/autoresearchinc/", "x_url": "https://x.com/hiloopai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/5b91efea37962386a163af58e2f2920b1e9d0aa7.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Ankit Gupta", "url": "https://www.ycombinator.com/people/ankit-gupta" }, "founders": [ { "user_id": 1714115, "full_name": "Karan Brar", "title": "Founder/CEO", "founder_bio": "Founder at Autoresearch, Inc. Previously ML at Reducto, Head of ML Infra at DynamoAI.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/brarkaran", "x_url": "https://x.com/deepmatmul", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/e596ed985a539262821de87d80ea418725ab7a90.jpg" }, { "user_id": 1597736, "full_name": "Thomas Boser", "title": "Founder/CTO", "founder_bio": "Founder at Autoresearch, Inc. Previously engineer at Reducto, founding engineer at Crosswise, senior ML engineer at Discord, ML at Sentropy & SoFi.", "is_active": true, "linkedin_url": "https://linkedin.com/in/tboser", "x_url": "https://x.com/thomasboser", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/7c1d2d3062a34b6d55d2d4e7aef55e887853400b.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 108337, "title": "hiloop: we run thousands of experiments to improve your models", "tagline": "Send us a hard task, your model or agent, and an eval. We run the research campaign and return a verified improvement.", "body": "We're Karan and Thomas, and we're building hiloop: **infrastructure for automated research**.\n\n**TL;DR: we gave two stock coding agents 50 B200s and our infra. They ran 4,188 experiments in two days and beat the published state of the art on [Karpathy's autoresearch benchmark](https://github.com/karpathy/autoresearch). The same machinery found a [NanoGPT speedrun](https://app.primeintellect.ai/speedrun/nanogpt) candidate that beats the best known result. [See](https://hiloop.ai/blog/search-is-enough/) full writeup.**\n\nNow we want to point it at your hardest problems.\n\n![uploaded image](/media/?type=post&id=108337&key=user_uploads/1714115/63396f7f-3e21-41ea-83ea-54cbaaa9f63f)\n\n**The results**\n\nOn Karpathy’s autoresearch benchmark, our final recipe reached **0.9016 val_bpb**. The previous best published result was [Recursive’s 0.9109](https://www.linkedin.com/pulse/first-steps-toward-automated-ai-research-recursive-si-kybtc) on B200 hardware. For NanoGPT, our implementation had **12.54% lower runtime** than upstream.\n\nThere was no bespoke research agent or elaborate scaffold. We used Claude and GPT-5.5 out of the box. What changed was the system around them.\n\n**What hiloop does**\n\nGive us a hard task, your current agent or model, and an evaluation criteria. Hiloop runs an autoresearch campaign across training, data, prompts, tools, harnesses, and systems, then returns the best verified improvement. It runs in your cloud or hosted by us.\n\nWe’re starting with agent and model training: SFT, post-training, continual learning, and optimization.\n\n[https://drive.google.com/file/d/1Rv8HttSHZ-yvSJk6hsWQsqCXWTHJQogK](https://drive.google.com/file/d/1Rv8HttSHZ-yvSJk6hsWQsqCXWTHJQogK/view?usp=sharing)\n\n**Who we are**\n\nWe met at Reducto, where we built its ML, post-training, and platform systems. Karan previously led ML infrastructure at DynamoAI . Thomas was a founding engineer at Crosswise and an MLE at Discord and SoFi.\n\n**Two asks**\n\n1. **Want your models or agents better, faster, or cheaper?** If you have a hard objective in post-training, continual learning, inference optimization, or classic ML, send us the task, your baseline, and the eval. We’ll personally run a campaign with you through verified improvement.\n2. **Building autoresearch or RSI loops?** Reach out to give our system a try! We’d love to compare notes.\n\nEmail [**founders@hiloop.ai**](mailto:founders@hiloop.ai) to get in touch with us!", "url": "https://www.ycombinator.com/launches/SBN-hiloop-we-run-thousands-of-experiments-to-improve-your-models", "votes": 8, "created_at": "2026-07-29T19:50:09.250Z" } ] }, { "id": 33639, "name": "Magma", "slug": "magma", "former_names": [], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/9b261522e21a27147c25d66eb0b126f8b62e5295.png", "website": "https://magmahq.ai", "all_locations": "San Francisco, CA, USA", "long_description": "We enable companies monetize their agents' traces.", "one_liner": "Monetize your agent's traces.", "team_size": 1, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1786606834, "tags": [ "Reinforcement Learning", "Data Engineering", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/magma", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/magma.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/magma-hq", "x_url": "", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/9b261522e21a27147c25d66eb0b126f8b62e5295.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Harshita Arora", "url": "https://www.ycombinator.com/people/harshita-arora" }, "founders": [ { "user_id": 3107607, "full_name": "Ramtin Shahzanian", "title": "Founder/CEO", "founder_bio": "failing my way through.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/ramtin-shahzanian/", "x_url": "https://x.com/42Ramtin", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/b58098b0fa3c418c52382d092d86f1801136d50d.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 110488, "title": "Magma - AI labs will pay you to learn from your agent's mistakes", "tagline": "We enable any company monetize its agents' traces!", "body": "**TL;DR:** Magma enables companies building AI agents to monetize their data in a few clicks.\n\n\n\n**❌ The problem**\n\nIf a product uses AI agents, every interaction creates valuable data. That could be a user correcting an agent’s mistake, or an agent discovering a new way to complete a workflow. That real-world data currently sits unused, while the best alternative is human annotators inventing hypothetical scenarios.\n\n**✅ What we do**\n\nMagma turns that data into recurring revenue for companies.\n\nCompanies share their agents’ traces in a few clicks. Access is read-only and encrypted, and nothing is shared without approval. Magma anonymizes the data, turns it into complex training environments, and handles licensing to AI labs. Every time the data is licensed, the company gets paid.\n\n**👋 Who I am**\n\nI’ve previously built the first Persian OCR model and legal LLM for my startup in Iran, formulated new transformer-based ML architectures in Oxford, and created a dev tool context management for agents that worked so well I decided to repurpose it to extract training data from traces (now Magma).\n\n![uploaded image](/media/?type=post&id=110488&key=user_uploads/3107607/07bc99c4-63f2-4235-870e-9b97977f8c43)\n\n**🧐 But why?**\n\nWe believe no two AI agents on the planet should be corrected for the same mistake. This is why we're building infrastructure to turn the experience of any AI agent into generalizable learning signals to improve any other AI agent.\n\n**🫡 Our ask**\n\nDo you know anyone at a frontier lab looking for RL environments built from real production workflows, or at an enterprise in need of fine-tuning their models on niche tasks? Let us know and we can help: [ramtin@magmahq.ai](mailto:ramtin@magmahq.ai)\n\nAlso, you can start monetizing your agents’ traces at: [magmahq.ai](http://magmahq.ai)", "url": "https://www.ycombinator.com/launches/Sk4-magma-ai-labs-will-pay-you-to-learn-from-your-agent-s-mistakes", "votes": 37, "created_at": "2026-08-14T01:00:00.832Z" } ] }, { "id": 33738, "name": "CoArena", "slug": "coarena", "former_names": [ "Coasty" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/1543623e6506eaf893b1c43e614688a6b2a32bda.png", "website": "http://coarena.ai", "all_locations": "San Francisco, CA, USA", "long_description": "CoArena is a live arena where anyone can use the world's top computer-use models racing two of them on the same real computer task and judging which one did it better. Every battle becomes something the AI labs can't build themselves, an honest test of their agents on real work, and the data to make them better.", "one_liner": "The biggest crowdsourced benchmark for Computer-Use", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Engineering, Product and Design", "launched_at": 1782294083, "tags": [ "AIOps", "Machine Learning", "Reinforcement Learning", "Data Labeling", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Engineering, Product and Design" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/coarena", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/coarena.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/coastyai", "x_url": "https://x.com/coastyai", "fb_url": null, "cb_url": "https://www.crunchbase.com/organization/coasty", "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/1543623e6506eaf893b1c43e614688a6b2a32bda.png", "app_video_url": null, "dday_video_url": null, "company_photos": [ "https://bookface-images.s3.us-west-2.amazonaws.com/attachments/cfec1f74a71f276cfe85bcc102647baf48c51fff.png" ], "primary_partner": { "name": "Brad Flora", "url": "https://www.ycombinator.com/people/brad-flora" }, "founders": [ { "user_id": 1194042, "full_name": "Prateek Jannu", "title": "Founder/CEO", "founder_bio": "Prateek Jannu, Co-founder & CEO - Stanford (MS. EE) and Purdue (BE. CE). Invented Solo-Synth-GAN, which generates labeled training data from a single example. Built Coasty to #1 on the OSWorld benchmark (82%), beating Agent S3, UiPath, Kimi K2.5, and Claude Sonnet 4.5. Now building CoArena.", "is_active": true, "linkedin_url": "https://linkedin.com/in/prateekjannu", "x_url": "https://x.com/prateekjannu", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/26ed45c60c34293a10373502e615a8c6499d356f.jpg" }, { "user_id": 2457255, "full_name": "Nitish Kovuru", "title": "Founder", "founder_bio": "Co-founder and CTO/COO of Coarena, launched Coasty prev, SOTA CUA framework on OSWorld at 83% accuracy. Did CS at Columbia (Vision track) and hold a BS in Computer Engineering from Purdue, where I TA'd a graduate-level AI course as an undergrad and later went on to build enterprise sales automation agentic systems for my previous company.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/nitish-kovuru-2423351b4", "x_url": "https://x.com/llmhubdev", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/29f588d1c332c655eee8ea69c5e1c27e9ec4fce9.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 111689, "title": "Coarena - The biggest crowdsourced benchmark for Computer-Use", "tagline": "We build the most advanced and biggest benchmark for computer use", "body": "We are officially launching CoArena (YC S26). Released 2 weeks ago, users have doubled every week since, and the arena went from $0 to $60,000 in revenue since. It's a live arena where the world's top Computer-Use models race to finish the same real computer task, and real people judge which one did it better, blind.\n\n\n\n![uploaded image](/media/?type=post&id=111689&key=user_uploads/1194042/43ddaa95-1f50-4b30-b0a4-9b085433ec3c)\n\n**The problem**\n\nEvery AI lab says their agent is the best at using a computer, and every claim points at a benchmark score. We know those scores well: earlier this year we built the #1 agent on OSWorld (82%). That's exactly how we learned the scores mean very little. Benchmarks are static, agents memorize them, and they say nothing about the messy tasks people actually need done.\n\n**What we built**\n\nOn CoArena, anyone can type a real task (\"find me the cheapest flight,\" \"assemble this expense report\"), watch two anonymous frontier agents race it live, and vote on the winner. Names are revealed only after you vote, so the judging is blind. Every battle becomes a fresh eval no model has ever seen, and the failure cases labs never catch: prompt injection, spending money when it should ask, silent side-effects.\n\n**What the arena has already shown**\n\n\\- Only \\~66% of agent runs finish their task at all 1 in 3 dies to a failure before any score exists\n\n\\- On identical tasks, the best frontier agent completes \\~87% of runs; the worst completes \\~42%\n\n\\- One major frontier model times out in roughly a third of its runs\n\n\\- In \\~7% of battles, human judges ruled that BOTH agents failed\n\n\\- The #1 spot on our leaderboard changed hands three times in three days, real tasks don't saturate\n\n**What you get**\n\nIt's completely free, including models you can't use anywhere else (Fable, Sol, and unreleased previews). Run your real errands with frontier agents, keep whatever they produce, and find out which model is actually best at YOUR kind of work, not the benchmark's.\n\n**Our asks**\n\n1\\. Use it, it's free. Bring your daily work: the flights, the spreadsheets, the research. Two frontier agents will race it, you keep the result, and your vote makes the leaderboard more honest: \n\n2\\. If you're an AI lab working on computer use or vision, or a robotics company building on vision, we'd love to talk. We can send you sample evals and data today: [founders@coasty.ai](mailto:founders@coasty.ai)", "url": "https://www.ycombinator.com/launches/T3R-coarena-the-biggest-crowdsourced-benchmark-for-computer-use", "votes": 59, "created_at": "2026-08-22T19:00:00.303Z" }, { "id": 105157, "title": "Coasty: The computer-use agent that doesn't break on real software", "tagline": "Coasty helps developers kill RPA.", "body": "**TL;DR: Coasty gives developers a computer-use agent that reliably automates back-office work in legacy desktop software where RPA scripts and other CUAs fail. It’s achieved 82.81% on OSWorld and it’s great at automating workflows where quiet mistakes cost the most: regulatory filings, patient records, payments, and gives developers precise control and observability over its work.**\n\n\\\n(_The portals and the data in it, are mockups_)\n\n**Replayable audit trail (real run):**  \n\n**OSWorld submission (Coasty scored 82.81% on OSWorld Verified, confirmed by the OSWorld team, with our latest internal run at 85.60%.):**[ https://github.com/coasty-ai/coasty-osworld](https://github.com/coasty-ai/coasty-osworld)\\\n\\\n**The Problem**\n\nMost back-office work doesn't happen in clean web apps. It happens in legacy desktop software, full of surprise pop-ups, wandering menus, and impenetrable UIs.\n\nToday, developers automating this back-office “real” work are stuck choosing between brittle RPA scripts and CUAs that demo beautifully and score high but break on contact.\n\nThe problem isn't getting an agent to click. It's trusting it when a silent mistake gets someone fired.\n\n### **Why Coasty is reliable where others aren't**\n\n**It recovers mid-task.** Coasty catches its errors and fixes them mid-run. If it’s crunching on a 20-step workflow and makes a misclick at step 3, it will recover and start from the error. First-pass accuracy is vanity. Recovery keeps a run alive.\n\n**It survives messy software.** Pop-ups, modals that steal focus, legacy desktop apps. Coasty doesn’t record clicks, it reads the screen. So when the UI changes overnight and every RPA script breaks, Coasty looks at the new screen and keeps going.\n\n**It knows when it's done.** An agent that thinks it finished but didn't is worse than one that fails loudly. Coasty verifies its own output and leaves a timestamped audit trail you can replay.\n\n**It swarms.** One task across forty state portals, each with its own ancient interface, all at once. Forty agents, one verified result set. What takes a team a week, is done overnight.\n\n### **Traction**\n\nCoasty has been verified by the OSWorld team and will be published soon, but also another internal benchmark we published (85.60%): < >.\n\nWe have paying customers on four continents, a few thousand active users, and we have actually grown 24x since Claude Cowork came out.\n\n### **Why us**\n\nWe started working on this idea when models weren’t good enough for computer use (last summer), so we’ve built frameworks to test and run the perfect combination of models and harnesses to see what works in practice. Prateek worked as an ML engineer at multiple startups and dropped his Stanford admit while I did CS at Columbia and built my previous company’s entire agentic sales automation. We’ve grown with this field and have been working on it even before it got much attention, so we understand where CUAs mess up and how to deal with these messes realistically.\n\n### **The ask**\n\nSend us one workflow your team runs by hand. We’ll tell you straight whether Coasty can run it reliably today. If it looks automatable, we’ll build a working proof of concept on your real software before a long sales process.\n\nWe’re especially looking for intros to operators and executives in accounting, tax, insurance, healthcare, compliance, and data privacy - anywhere teams are still babysitting RPAs, clicking through legacy portals, reconciling records, filing forms, or moving data across old desktop software.\n\nHere's our Calendly:[ https://cal.com/coasty/15min](https://cal.com/coasty/15min)", "url": "https://www.ycombinator.com/launches/RM5-coasty-the-computer-use-agent-that-doesn-t-break-on-real-software", "votes": 12, "created_at": "2026-07-10T17:30:25.886Z" } ] }, { "id": 33806, "name": "Experiential Labs", "slug": "experiential-labs", "former_names": [ "Resolute Labs" ], "small_logo_thumb_url": "https://bookface-images.s3.amazonaws.com/small_logos/716e828c9efd6efe6da64574ff67dd5bfa0c2a25.png", "website": "https://experientiallabs.ai", "all_locations": "San Francisco, CA, USA", "long_description": "We are building an open source gateway for AI budget holders to control model access and turn their traffic into better models that they own. We're a team that has earned an ML PhD at Georgia Tech, scaled AutoGPT to 160k GitHub stars, and led a self-driving simulation team at Waabi.", "one_liner": "Open source AI gateway that turns your traffic into better models", "team_size": 2, "industry": "B2B", "subindustry": "B2B -> Infrastructure", "launched_at": 1782607347, "tags": [ "Reinforcement Learning", "Infrastructure", "AI" ], "tags_highlighted": [], "top_company": false, "isHiring": false, "nonprofit": false, "batch": "Summer 2026", "status": "Active", "industries": [ "B2B", "Infrastructure" ], "regions": [ "United States of America", "America / Canada" ], "stage": "Early", "app_video_public": false, "demo_day_video_public": false, "app_answers": null, "question_answers": null, "url": "https://www.ycombinator.com/companies/experiential-labs", "api": "https://devasheeshg.github.io/yc-api/batches/summer-2026/experiential-labs.json", "year_founded": 2026, "city": "San Francisco", "country": "US", "linkedin_url": "https://www.linkedin.com/company/experiential-labs/", "x_url": "https://x.com/explabsai", "fb_url": null, "cb_url": null, "github_url": null, "logo_url": "https://bookface-images.s3.amazonaws.com/small_logos/716e828c9efd6efe6da64574ff67dd5bfa0c2a25.png", "app_video_url": null, "dday_video_url": null, "company_photos": [], "primary_partner": { "name": "Diana Hu", "url": "https://www.ycombinator.com/people/diana-hu" }, "founders": [ { "user_id": 3149739, "full_name": "Kion Fallah", "title": "Founder/CEO", "founder_bio": "Co-founder and CEO of Experiential Labs. Previously Staff Research Scientist at Waabi leading the mixed reality simulation team. PhD in Machine Learning from Georgia Tech focused on synthetic data.", "is_active": true, "linkedin_url": "https://www.linkedin.com/in/kfallah", "x_url": "https://x.com/okfallah", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/eb099f2fb8bbfee19779eae1dae81307ea85d07d.jpg" }, { "user_id": 411375, "full_name": "Silen Naihin", "title": "Founder/CTO", "founder_bio": "Built and grew AutoGPT to 160k stars. Did AI for science at the DOE. Previously cofounded Stackwise (was in W24) and various founding AI eng roles. Did continual learning, interp, and evals research.", "is_active": true, "linkedin_url": "https://linkedin.com/in/silen-naihin", "x_url": "https://twitter.com/silennai", "avatar_thumb_url": "https://bookface-images.s3.us-west-2.amazonaws.com/avatars/b68b10165c34092f3bb0c78caccfa5794a625212.jpg" } ], "jobs": [], "news": [], "launches": [ { "id": 109456, "title": "Experiential Labs - Turn your AI spend into a cheaper + better model you own", "tagline": "Train a model your company owns, guaranteed to match or beat frontier quality at 50% lower cost.", "body": "**TL;DR: Companies are spending more on AI every month, but almost none of that spending becomes an asset they own. We train a small language model for your company's use cases. You fully own it, and we guarantee the same or better quality as the frontier models you use today at 50% lower cost.**\n\n\n\nCompanies are spending more on AI every month, but almost none of that spending becomes an asset they own.\n\nThe prompts, traces, and use cases are yours, _but the model stays rented._\n\nSo you keep paying frontier prices for repetitive work a smaller, specialized model could do faster, cheaper, and often better\n\nWe built Experiential Labs to change that.\n\n**Here's how it works**\n\nKeep using GPT, Claude, or whatever models are already in your stack.\n\nExperiential Labs uses your traffic to build a simulation of the tasks, failures, and edge cases your agent encounters. You can also upload existing traces to bootstrap the process.\n\nThe platform trains you a model against that simulation, so it gets better at your specific use cases over time.\n\nAs your model improves, it earns more traffic. Hard requests can still go to the frontier. Everything else moves to your faster, cheaper model.\n\nYou own the weights, can run them wherever you want, and keep the profits they create. When a stronger base model comes out, we transfer what your model has learned instead of making you start over.\n\nOn some tasks we've seen up to 97% lower cost and +50% higher quality. Across a full deployment, we guarantee at least 50% lower cost at the same or better quality, backed by an SLA.\n\n**The team**\n\nWe are both AI researchers. Kion built simulation systems for autonomous vehicles at Waabi after earning his ML PhD at Georgia Tech. Silen built and grew AutoGPT to 160,000 GitHub stars and has worked across continual learning, evaluations, interpretability, and AI for science.\n\n## Our ask\n\nIf your company sends significant traffic to GPT, Claude, or other frontier models, we would love to help.\n\nIntros to founders and AI leaders with growing inference bills are also greatly appreciated.\n\n**Email:** [founders@xplabs.ai](mailto:founders@xplabs.ai) \n\n**Schedule a call:** \n\n**Website:** [xplabs.ai](http://xplabs.ai) ", "url": "https://www.ycombinator.com/launches/STQ-experiential-labs-turn-your-ai-spend-into-a-cheaper-better-model-you-own", "votes": 18, "created_at": "2026-08-06T22:00:00.114Z" } ] } ]