hillclimb
Training Data for Recursive Self-Improvement
Y CombinatorFall 2025ActiveB2BEngineering, Product and DesignReinforcement Learning
What they do
Problem
Frontier AI labs lack high-quality training data and workflows to teach AI agents to iteratively improve and perform AI research tasks like an AI research scientist.
Solution
They provide training data and related tooling to train agents for recursive self-improvement toward AI research scientist capabilities.
Who it's for
Frontier AI labs developing advanced AI agents and research systems.
Founders
Jun Park
Founder
hillclimb / prev deepmind, pro valorant, georgia tech
Open roles 4
Company timeline
- Founded2025
- YC batchFall 2025
- StatusActive
- Team size4 people
- LocationSan Francisco, CA, USA
