Home›Companies›Repovive
Repovive

Scaling AI formal verification for chips and critical software

We work with AI labs to improve models on formal verification. We have $100k in contracts across two formalization RL environments and a $200k LOI for our broader coding data. Our team includes Olympiad medalists, former members of frontier labs, PhDs, and former Google engineers. We’ve published papers at leading AI conferences, including NeurIPS, ICLR, and IJCAI. We’ve formed the densest community cluster of 12,000 IMO, IOI, and ICPC medalists.
Active Founders
CS PhD at UMD. National Informatics Olympiad gold medalist. Published papers at leading AI conferences, including NeurIPS and ICLR. Built math datasets with Lean-verified proofs for AI labs. ICPC World Finalist, and ICPC North American Championship bronze medalist. International Grandmaster on Codeforces, ranked among the top 200 worldwide (top 0.1%). Built the Algorithms with Shayan YouTube channel with an IOI audience of 33,000 subscribers and 1M+ views.
Amir Mostafavi
Amir Mostafavi
Founder/CPO
Co-founder @ Repovive. Ex-Google and YouTube. Leading teams at Google Home (post-training Gemini for Home and Home experiences) and YouTube Shorts. MS in Human-Computer Interaction and BS in Computer Science from UT Austin.
Company Launches
Repovive: Enabling AI to formally verify chips and critical software
See original launch post

Hey YC! We’re Shayan and Amir, co-founders of Repovive. We’re enabling AI models to formally verify the chips and code the world runs on.

A flaw in a chip design can be expensive to fix after fabrication. A bug in critical software can expose the infrastructure we depend on. Formal proofs can establish that a system meets its specifications, but constructing them takes specialized expertise and substantial manual work. Teams often rely on tests that cover selected scenarios, leaving gaps where failures can go undetected.

https://www.youtube.com/watch?v=CL_-F3LWi34

Today, we provide AI labs with formalization data and reinforcement learning environments in Lean for model training and evaluation. Lean is a proof assistant for writing and checking mathematical proofs. Our non-public Lean tasks come with formally verified reference proofs, giving labs difficult problems for training and a rigorous way to evaluate whether models can solve them.

We’re already working with AI labs and have closed $100k in contracts in just the past month. Our team includes Olympiad medalists, PhD researchers, and former Google engineers, with research published at NeurIPS, ICLR, and IJCAI.

This work lays the foundation for our larger ambition, a black box for formal verification. You specify what needs to be proven, and the system does the work of finding a proof. Any proof it returns can be independently and formally verified. Our goal is to make formal verification practical enough to become a standard step in chip design, critical software development, and mathematical research.

We’d love introductions to

  • AI researchers and post-training teams sourcing formalization data or reinforcement learning environments.
  • Engineers and researchers tackling difficult verification problems in chip design, critical software, or mathematics.

We’re happy to share samples of our Lean tasks and learn more about what your team is working on. Reach us at founders@repovive.com.

Repovive
Batch:Fall 2026
Team Size:5
Status:
Active
Location:San Francisco
Primary Partner:Christopher Golda