I build agentic AI for regulated domains — healthcare claims, cross-border compliance, financial risk — where a wrong answer has a real cost and "it usually works" isn't good enough.
Two planes — policy and runtime — and the verification tooling that keeps them honest. AI interprets; deterministic engines execute; every decision leaves a trace.
A regulatory rule-execution engine. Compiles regulation into content-addressed, dual-signature-verified artifacts; a Rust kernel executes them deterministically and fails closed on a verification miss.
Cross-border digital-asset compliance, evaluated in real time. Resolves jurisdiction conflicts across MiCA, FCA, and the GENIUS Act, and moves a position through explicit verdict states: compliant → conditional → blocked.
A Claude Code plugin that holds agents to their evidence: blocks unverified claims, refuses silent git-history rewrites, and enforces a hard line between what's implemented and what's merely planned.
A fitness-gate: an irreversible action is authorized only by a reproducible, measured demonstration that it meets declared criteria. No valid demonstration means denial — by construction.
Upstream label correction for multi-omics cohorts. Detects mislabeled samples through cross-omics concordance scoring, before bad labels poison everything downstream.
A lending default-detection system — quantitative risk modeling in the same auditable mold: every score traceable to the inputs and the model version that produced it.