I build practical tools and software experiments with AI coding agents. My work centers on automation, testing, debugging, and making results easy to inspect.
These are independent, substantially AI-assisted projects. I guide the work through planning, prompting, coding, testing, and debugging.
| Project | What you can inspect |
|---|---|
| Clearfile workflow demo · Python / SQLite | A local quote-to-invoice demonstration with decimal arithmetic, safe retries, transactional numbering, and approval checks. Synthetic data throughout. Live walkthrough. |
| Butter Lab · JavaScript / WebGPU | An interactive heat-and-flow simulation with reference-model checks, GPU health probes, and documented modeling limits. Try the lab. |
| Recorded control harness · Node.js / Playwright / Python | Bounded actions, request receipts, recording-health gates, and a run audit. Methods and limitations. |
| New Vegas controller · Python | Typed API decisions, conservative budget reservations, telemetry, independent watchdogs, and regression cases from a supervised campaign. Architecture. |
- Super Peasant for Fallout 4 — native C++ and Papyrus companion mod, with a playable preview, runtime-specific validation, and credited third-party assets.
- Left 4 Dead 2 controller — asynchronous decisions, a VScript observation addon, capture controls, and regression cases.
- Geometry Dash experiment — a native bridge and Python controllers with state-checked input and recording acknowledgements.
Start with a concrete task. Build a small working path, inspect failures, add useful checks, and document what the result supports.
The game experiments use supervision and direct game state; some pause the game while planning. Each repository explains its setup, methods, and limits. The workflow demo uses fictional business data, and Butter Lab is an educational approximation.