I build AI-assisted data systems for large, messy, high-stakes datasets, then turn the results into usable software, analysis, and plain-English insight.
- Evidence databases for massive document dumps — CaseStack, an engine that turns any document dump into a quarriable evidence database. Its proof-of-concept deployment over 1.38M DOJ documents draws ~15K daily users, and its results have been used by reporters at major and independent outlets. Data exports: Epstein-research-data
- Scientific extraction pipelines with provenance — Organoid Protocol Atlas, an ongoing effort, is an interface grown from a curated set of academic literature on organoid culture protocols.
- Bioinformatics tools and high-performance search — AmpliconHunter2, a SIMD-accelerated in silico PCR engine, 5.9× faster than the original (ICCABS 2026)
- AI-agent workflows for research and software engineering — claude-code-scientist, a 24-skill semi-autonomous research agent with validation hooks and provenance tracking for academic literature acquisition, data acquisition, computational experimentation, synthesis, and peer review. "Self-improving": includes /cortex, which audits prior sessions, diagnoses issues, and generates fixes.
I scaffold AI agents the way you'd onboard a new hire: they work in Git, write findings up as reports, and pass a peer-review check before the output reaches me. claude-code-scientist shows one example of this setup — staged skills, validation hooks, and an automated review pass.
Resume · LinkedIn · Google Scholar · ORCID
AmpliconHunter1 · AmpliconHunter2 · Epstein-data.com · CaseStack · OrganoidProtocolAtlas