I turned a quality obsession into a governed AI engineering org.
Senior quality engineer, seven years. In an AI era the scarce skill isn't writing tests — it's evaluation: defining what “good” means, building the harness that measures it, and turning that into a release decision you can trust. I ship that as real software — a release-confidence engine on npm and a graded QA-practice platform — not slideware.
Voice is “we”. This site is “I”.
QA taught me the real question is never “does it pass?” — it's “how confident are we, and where's the risk?” So I build the evaluation systems that answer it: the rubric, the harness, the score, and the honest verdict. Each idea below is something I've shipped.
A green suite means “the cases I wrote pass” — not “ship it.” qulib fuses live-app evidence, automation maturity, and API coverage into one scored verdict — ship · caution · hold · block — and refuses to fake confidence it hasn't earned.
On NotQuality.com free-text bug reports are graded against a documented registry — coverage, severity, repro quality — the same way you'd score an LLM's output. Designing that answer key is the eval-engineering skill.
AI is great at finding unknown gaps — but the moment a judgment matters, it needs a reproducible check behind it. I draw that line on purpose: deterministic gates scale, the model reasons over real signal, and every verdict is one you can re-run and verify yourself.
Three things in flight right now — full detail on each lives in selected work below.
Not a mockup. Outer-loop principal agents — Route, Check, Fetch, Scan, and Build — govern policy and learning while inner-loop execution agents run inside project workflows. Press run and those five principals execute a real release-readiness sweep against the live notquality.com deployment: an actual HTTP fetch, a transport + security-header audit, and a11y/seo signals parsed from the page — rolled up into a release-confidence score you can verify yourself.
Scholastic, Inc. · New York, NY
Blue Apron · New York, NY
Want to work together, collaborate, or just talk systems? Send a note. For client engagements, head to tapquality.ai. I reply to most things within a day.
Looking to hire? Skip the back-and-forth.
start an engagement ↗