Writing, building, thinking.
Things I've built to think with.
PIR Generator
→Paste a raw Slack incident thread, get a complete Post-Incident Review. A 4-agent pipeline — Extractor, Analyst, Writer, Verifier — that reconstructs timelines, identifies root causes, and checks its own output for hallucinations.
TPM Interview Coach
→Record your answer to a TPM interview question and get an instant scorecard: WPM, filler words, CART structure, sentence flow, and an AI critique that tells you whether your answer sounds senior.
Identity Transaction Explorer
→Trace a complete identity-provider transaction through request validation, session resolution, policy evaluation, claim minimization, and session governance.
FlakyTestHunter
→Load a CI test suite's run history and instantly see which tests are actually lying to you. A real inconsistency-scoring algorithm — entropy plus transition rate, confidence-weighted by sample size — ranks every test by flakiness, then a one-click quarantine toggle shows what the worst offenders were doing to your real pass rate.
Zero-Server Secret Redactor
→A neural network that trains and runs entirely in your browser via TensorFlow.js, learning to spot and redact API keys, tokens, and connection strings before they ever leave the client. Teach it a new secret pattern live and watch it generalize — zero network calls, zero server logs.
Aether
→Benchmarks Transformers.js v3 running client-side via WebAssembly and WebGPU against a persistent Node.js server backend — a one-click test suite that isolates exactly where edge inference beats the cloud round-trip.
Security Lab.
Thinking out loud.
I move across functions and org boundaries for a living. The work that interests me sits in the seams — between teams, between disciplines, between the plan and what actually gets shipped. Lately that means AI systems, the messy human processes around them, and writing down what I learn so I'm forced to actually understand it.