The work, in motion.

A face-swap video generator: pick a GIF, add a face, get a shareable clip — with a GPU fleet, queue, and spend model built to survive real traffic.

RDS restore drill (recovery time) 7m 25s · 2026-04-18

A voice-native cooking assistant that reads a recipe aloud and guides you hands-free, step by step, in real time.

Automated tests passing 277 backend + 120 frontend · 2026-07-02

A local multiplayer party-game platform: a Godot host app renders the game on the TV, phones become controllers, and one wire protocol keeps a game engine and a native app in sync.

Automated tests passing across four runtimes 158 backend + 66 website + 32 controller + 23 Godot host · 2026-07-04

A writer's platform that turns a manuscript into data-driven story analysis: character dossiers, timelines, and structure maps.

Automated tests passing 621 backend + 240 frontend · 2026-07-03

A podcast discovery app built on one unified graph: live events, the host-and-guest relationship web, and the merch those people sell, each enriching the others.

Podcast episodes ingested and graphed 5,500+ · 2026-04

A family movie-night assistant that tracks each member's success and failure sentiment on every pick, so 'what should we watch' stops being a fight.

Recommendation endpoint latency (Gemini + OMDb), cold vs. cached 34.7s cold to 0.24s cached · 2026-07-03

A Discord-native game-night orchestrator that finds what a friend group can play together, then gets them from 'let's play something' to launched.

Steam library imported (end-to-end proof run) 182 games · 2026-06-12

A subscription personal-shopping assistant designed security-first: three privacy zones, one that must never leak a surprise gift to its own recipient, specified in locked decision records before a line of product code existed.

Design docs authored before the build began 199 · 2026-05-21

An autonomous agent that triages user feedback across 8 products, filing issues and dispatching fix PRs behind a human-approval gate.

Products served by one service 8 · 2026-06-06

A production multi-lane LLM job system that non-technical teammates run by dropping files into a shared folder, no client and no dashboard, just a queue that behaves.

Supervisor size (stdlib only, no framework) ~720 lines · 2025

An offline speech pipeline, Whisper transcription plus Kokoro voice synthesis, that replaced a discontinued paid captioning and voiceover vendor with zero cloud dependency.

Two Model Context Protocol servers that let an LLM operate After Effects and Premiere Pro directly, one validated tool call at a time.

Tools validated against a live Premiere Pro session 43 of 97 defined · 2025

A chief-of-staff agent that runs entirely as markdown read by the model: self-installing, self-repairing, versioned along the org chart, and adopted by colleagues inside a Fortune 500.

Release cycles maintained 4+ · 2025

This site: a static Astro portfolio engineered to be read by AI screeners as deliberately as it is read by people, built by a fleet of AI agents under a human-gated review process.

Machine-readable surfaces shipped alongside the human-facing pages 5 · 2026-07-04