I study how to evaluate and supervise AI agents, and build tools that make experiments easier to inspect and reproduce.
From Trinidad & Tobago. Studying Artificial Intelligence (Intelligent Systems) at XJTLU.
When every agent fails a task, how do we distinguish genuine difficulty from a broken task or a problem with the evaluation?
NeurIPS 2026 Verify-Agents workshop: poster accepted. AAMAS 2027: submitted, not accepted.
An open-source tool that keeps evaluation scores with their configuration, run artifacts, and file hashes so the record can be checked.
Open source · reproducible synthetic demo