Selected work
Case studies from my work, each short enough to read in two minutes. Every one follows the same shape: the problem, what I did, what happened, and what I'd tell someone else.
Measuring whether an AI quality system could be trusted
Built a 20,000-line evaluation harness for a GMP AI system: top-20 retrieval went from about 3-in-10 to about 8-in-10, and the unsafe recommendations I'd seen went from 3 to 0 across 201 test cases.
Two textbook RAG improvements that didn't earn their keep
Cross-encoder reranking and hybrid search are the standard next steps for retrieval. In the configs I tried, one scored worse and was far too slow, the other showed no measurable difference, and the wins came from boring changes.
Real-time telemetry from instrument to browser
Designed and built the message-processing subsystem that streams live instrument data from RabbitMQ to the browser without swamping the client.
Six years of software under 21 CFR Part 11
A thread rather than a single project: role models, audit trails, protected-action logging, and tamper-evident evidence, built into GMP software over six years.