Review STEM problems and simulated research environments for quality as training data. You will evaluate task prompts, ground truth answers, grading criteria, and model rollout transcripts across two review types:
Project A — Simulated Research Review: Assess simulated-research environments (seed + 3+ agent trajectories) for scientific realism, construction quality, and diversity.
Project B — STEM QA Review: Evaluate individual STEM problems for task clarity, ground truth correctness, grading soundness, and training data suitability. Verdict: Accept / Revise / Reject.
MS or PhD in Electrical Engineering or closely related field
Ability to independently verify ground truth solutions for grad-level EE problems
Strong scientific reasoning and written communication
~2.5 reviews per week (~2-4 hours per review)
3-week engagement (6/5–6/25)
Read complex STEM problem prompts and verify golden answers
Evaluate grading methodology and robustness
Read and summarize model agent trajectories (best and worst rollouts)
Provide structured feedback via standardized review forms
Render Accept/Revise/Reject verdicts with concrete explanations
Work is done on HFI (Surveyor)
48-hour lease per problem, extendable
Self-select problems within your domain expertise