Every system below is real, and every one is labeled with its true status — live in production, in active build, a working prototype, or anonymized client work. No mockups dressed up as launches. This is the range you get when you hire the person who also runs the QA.
Shipped and running in production.
A transparent trading-media platform: real-time market data, a signal engine, a research desk, and a public track record — the "printing press with the lights on".
A markets and AI-engineering academy: structured courses with in-browser code labs, mastery quizzes, certificates, and a retrieval-grounded AI tutor.
The agency site you are on right now runs its own QA: a proposal→deposit→client-portal money path, an operator CRM, and a CI proof gate (unit + a11y + Lighthouse) on every change.
A programmatic short-form video pipeline — script to rendered vertical video — for a $0-marginal-cost content channel plus a paid generative lane.
Quantitative research infrastructure over a market-data lake: rule-based setup detection, meta-labeling, and combinatorial purged cross-validation before anything is trusted.
Real, working, still hardening toward GA.
A cross-platform mobile app built to production discipline — every route hardened against missing params — with a four-tier verification harness gating each build.
An AI-driven social app with a custom quality engine on the backend, shipping to iOS through an automated EAS / TestFlight pipeline.
A daily-discipline app with a shared web + mobile core; the web app has completed its backend cutover behind a fixture-mode toggle.
Designed and specced; selectively built.
An AI car-diagnostic concept: photo and symptom input into a guided triage flow. Full product blueprint and mobile hero flow designed; build is selective.
An AI assistant that helps people appeal insurance denials — turning a denial letter into a structured, evidence-backed appeal. Specced and documented; gated on domain validation.
Delivered for a client — shown anonymized.
An AI receptionist and operations app for a home-services business: voice intake, photo-based estimates, and a lead-to-job pipeline. Delivered under NDA — shown anonymized.
A dark, high-motion commerce and content site for a creator brand — editorial art direction with a production-grade component system. Shown anonymized.
I take on a small number of AI build and QA engagements. If you have something real to ship, let’s scope it.
A demo shows five happy-path cases; production throws ten thousand real ones.
How to grade thousands of AI answers a day — a rubric plus a calibrated judge model.
Three words people confuse — and why you need all three for AI.
Scoring the retrieval itself: precision, faithfulness, and citation accuracy.
One prompt tweak silently breaks five other things — the fix is a golden set you re-run.
A hallucination is the model doing its job — fluent and confident. You gate for it.
Your code has a pipeline; your AI ships on "looks fine." CI for AI runs evals on every change.
Accuracy hides the failures that hurt — the number that matters is the failure rate on what matters.