AI Safety & Student Researcher @ UC Berkeley
I build statistical checks for AI evaluations, deterministic verification gates around autonomous agents, and empirical tools in computational biology — keeping negative results, calibration intervals, and abstentions in plain sight.
- 🦦 Core question: What does the evidence actually support once a model, a genetic screen, or a sensor hands back a confident answer?
- 🏆 Recent highlight: BlackBox won the Grand Prize at the 2026 Build with DataHub Agent Hackathon (600+ entries).
- 🔬 Open source: Contributing evals, power analysis, and genomics fixes to inspect_evals, control-arena, inspect_petri, and scverse/anndata.
more builds & live demos: faithful · night-orders · tideline · model-citizen · dosedna · cowboys-vs-wizards





