Reinforcement Learning at Berkeley
Berkeley's applied RL lab. Run by students, funded by industry.
RL environments and expert data for the labs training foundation models. Built by every major on campus.
Founding cohort · Fall 2026 · Open to every major
Built in the formats labs already use
What we do
Two departments. One loop.
Research trains the agent. Engineering builds the environment it learns in.
one loop · two departments
Open to every major
Every major has a dataset only it can label.
Builders make it run. Experts make it worth paying for. Tap a major.
Philosophy
Argument rubrics and ethics-case graders
How it works
One club, two doors.
Companies buy the pod. Students join it.
One vendor for the environment and the experts who write its tasks.
Two things to buy
Environments and expert data. Same pod, same invoice.
Second-expert review
Every label reviewed. Rejection rate is the metric.
Fixed scope, fixed price
One proposal, one IP agreement per job.
The unit of work
Every engagement ships from a pod.
Builders make it run. Experts make it worth paying for.
1 Board lead
Scope and delivery
2 Builders
Harness and verifiers
4 Domain experts
Tasks, rubrics, gold answers
One pod · one QA process · one invoice