AI Agents

Treasury Reapers: Sentient Arena Challenge 0 final report

Final report for my Treasury Reapers team submission to Sentient Arena Challenge 0: a goose-based OfficeQA agent, prompt and skill-contract iterations, leaderboard progression, and lessons from trace-driven optimization.

Porting scBench to inspect_evals: lessons from running AI agents on single-cell data

Lessons from moving scBench’s public task set into the inspect_evals framework, including scoring rules, evaluation setup, and what run logs revealed.