15 years in industrial quality. Building practical software with coding agents.

I define the work, constraints, and acceptance criteria. Coding agents implement and verify it.

A Team — Multi-agent infrastructure for AI coding assistants.
26 specialist agents, 19 enforced workflow skills, a lead orchestrator, and a Pipeline Auditor that checks whether work was actually done — not just reported.
One folder. Zero configuration. Works with Claude Code, Codex CLI, Cursor, and OpenCode.

builder.* — Domain packs that extend A Team into specific areas:

• builder-ai — Evals, RAG quality, prompt versioning, and safety gates
• builder-design — AI UI states, streaming, accessibility, and design tokens
• builder-product — PRD quality, metrics, and research synthesis
• builder-growth — Positioning, copy, and experiment discipline

Quality-Engineering-Skills — Open-source skills and agents for quality engineers working with AI, grounded in real industrial quality practice.

Agent HUD — A quiet interface for Raven Prism smart glasses that shows when a coding agent needs a decision, without demanding constant attention.

DOOMed Prism — An experimental DOOM port for Raven Prism smart glasses, built around a Qt-composited shared-memory framebuffer and gaze-driven controls. The Raven team compiled and ran it on actual Prism hardware directly from the repository.

These are experiments in wearable interfaces, unconventional input, and what becomes possible when software is designed around smart glasses rather than adapted from a phone.

TIO — A personal assistant that runs locally, keeping data off the cloud where practical.

KidOS — An Android app for young children where every session ends with an offline activity. No internet, gamification, or streaks.

Cody — A small tool for ICD-10 clinical coding.

Personal projects. Work in progress.

Python · C · FastAPI · PySide6 · React · TypeScript · PostgreSQL
Claude Code · Codex · agentic development workflows

Questions or issues → open one in the relevant repository.