Staff backend engineer, 14 years in. I build and run the platforms that other teams depend on: hundreds of millions of records, and a lot of people asking "is it down?" I'm the one who finds the root cause, at whatever layer it's hiding, and then makes sure it stays fixed.
Most of what's public here is a deliberate practice in directing AI coding agents well. Architecture decision records are the guardrails, issues are the spec, CI is the quality gate, and I act as architect and reviewer.
Things that work
- woodwire: secure, event-driven chat between a static PWA and a fully-local AI agent. Zero-trust edge, SQS/S3 message flow, no open ports. I use it every day.
- whassat (repo): fully offline object detection in the browser, in under 6MB. Built by hand in 2019, when my kid started pointing at everything and asking what it was.
- Blueprint templates: a family of GitHub template repos that turn lessons learned into paved roads, from "use ADRs" up to an empty repo → a deployed AWS data pipeline.
- Commit Hygiene: why and how I curate commits and pull requests.
Works in progress: hoopstat-haus (an NBA/WNBA stats lakehouse; the pipeline code is built and tested, but it isn't publishing data yet) and scrymat (a peer-to-peer Magic: The Gathering playmat). Each README has an honest status section.
Earlier: reactifex (~660k npm downloads, part of Open edX's frontend platform) and a few hundred merged PRs across Open edX.