Hey all, we just launched Creeks ( https://creeks.ai ). We build and sell training data, eval data and RL environments for AI teams. Three domains for now:
Code: multi-file tasks in real repos with tests, agentic coding trajectories, held-out eval sets
Clinical: diagnostic reasoning and differentials, grading patient-facing answers, clinical note summarization and extraction
Education: tutoring dialogues that guide the student instead of handing over the answer, worked solutions, grading rubrics for essays and proofs
The whole thing depends on two things: the data is made by people who actually do this work, and the outcomes are checkable. Tests where they exist, clear rubrics where they don't.
If you're at a frontier lab, a neolab, or a team fine-tuning your own models and need data in any of these areas, tell me what you're trying to improve and we can start with a small pilot.
If you're a software engineer, clinician or educator, you can help us curate this data and get paid for it.…