liteOdyssey
A rare-disease diagnostic system whose reasoning and tool use can be inspected.
Clinician-gated policy · held-out evaluation · blinded physician review
Patient phenotype↓Policy + tools↓Ranked differential
Research portfolio
Build the system.
Inspect the evidence.
Scientific agents, written policies, and experiments that test when their answers hold up.
I design and implement agent workflows, tool environments, and the controls that keep evaluation interpretable.
PIHF-MCP
An agent that investigates model failures and tests changes to a written policy.
Fixed model weights · recorded development runs · pinned benchmark scorers
Investigate failures↓Revise + test policy↓Keep or roll back
Hair-loss evidence + access
A bilingual project bringing together medication-safety research and selected individual hair-loss outcomes.
Population evidence and individual observations, with their limits kept explicit
ResearchSelected outcomesEnglish + Tiếng Việt
The common thread
A score is an estimate. I bring experimental design and epidemiology to the question of whether a result deserves our confidence.
Research background ↗