Reinforcement Learning
2 posts
Closing the Loop: Turning Live Production Data into RL Training Signal
Static training sets go stale the moment a model ships. Notes on building a flywheel that converts real production failures into new RL training data and gyms.
Jul 8, 2026·3 min read
Building Reusable RL Environments for Coding Agents
Notes on architecting software-engineering RL environments ('gyms') that scale from a handful of hand-built tasks to 100+ environments authored by a whole organization.
Mar 10, 2026·4 min read