I build useful systems,
then make them run reliably.

Public source

A privacy-first Chrome extension that scores and labels likely AI-generated images directly in the browser. The quantized vision model, metadata checks, and evidence UI run locally with WebAssembly—no uploads, backend, or API key.

Public source

A small production platform for the tools I actually want to use. Multiple Dockerized applications run behind a reverse proxy on my VPS, while a deployment daemon watches their GitHub repositories and rebuilds changed services automatically.

Public source

A custom Gymnasium environment for training a pursuer drone to intercept a moving target in 3D. It includes camera-like observations, target-lock memory, fly-through collision geometry, shaped rewards, PPO training, and a lightweight renderer.

Public source

A deliberately unserious vision app that audits screenshots of coding setups, scores the visible agent and tool sprawl against a structured rubric, and publishes shareable results to a public leaderboard.

Private build

A private tooling pipeline that turns repeatable browser interactions into composed product-demo videos. I built it to spend less time manually laying out and recording every launch clip—and more time improving the products themselves.

Work project

A human-in-the-loop support automation that combines deterministic browser work with an LLM-powered Azure DevOps search. Selenium extracts failure context; Claude Code reasons over candidate bugs; a strict contract returns parseable JSON.