LeanAgent: Lifelong Learning for Formal Theorem Proving

ℹ️ EPISODE INFO

Released

August 6, 2026

Duration

6m

Channel

YC Root Access

Watch on YouTube

▶ Open ↗

📝 EPISODE DESCRIPTION

At our inaugural YCML at Startup School, YC Partner Ankit Gupta speaks with Adarsh Kumarappan about LeanAgent, a continual learning framework for proving formal mathematics in Lean. LeanAgent builds a curriculum from Lean repositories, learns to retrieve relevant mathematical premises, and uses tree search to construct proofs. The goal is to let a theorem-proving system learn new mathematics without forgetting what it already knows. In the paper, LeanAgent produced 155 new formal proofs across 23 domains and demonstrated backward transfer, where learning new subjects also improved its performance on earlier ones. Apply to Y Combinator: https://www.ycombinator.com/apply Work at a startup: https://www.ycombinator.com/jobs

🎤 SPEAKERS

• Ankit Gupta

  host

  Host/interviewer for YC Root Access (Y Combinator).

• Adarsh Kumarappan

  guest

  Researcher presenting LeanAgent, a lifelong/continual learning approach for formal theorem proving in Lean.

🧾 EPISODE SUMMARY

In this episode of YC Root Access, featuring Ankit Gupta and Adarsh Kumarappan, LeanAgent: Lifelong Learning for Formal Theorem Proving explores leanAgent uses lifelong learning to improve automated Lean theorem proving LeanAgent frames formal theorem proving in Lean as a lifelong learning problem, balancing stability (not forgetting) and plasticity (learning new material).

🔗 RELATED EPISODES

Evaluating the Fine-Grained Planning Abilities of Web Agents

ChartNet: Training Vision-Language Models to Understand Charts

Any-Horizon Reasoning for Video Agents

Zero-Shot Predictive Models for Relational Databases

Improving Small Language Model Reasoning With A* Search

Interpretability and Safety for Robot Foundation Models