Projects/
Evolves AI models the way biology breeds organisms — random variation, selection, repeat. Built to run on a single GPU, fast enough that the approach actually competes with modern deep learning.
A simple strategy game — think Risk with planets — that doubles as a training ground for AI agents learning long-horizon planning.
Automates the back-office of running rental properties. Booking emails get parsed, contracts auto-generated, calendars synced — built on top of the Google spreadsheet + email ecosystem.
An attempt to rediscover the laws of light by evolution. Simulated particles wander; an algorithm searches for force rules that make them behave like real light through a slit.
The configuration that runs my mini-datacenter at home — a dozen self-hosted services, all reachable through one private network, automatically reproducible if anything breaks.
Lets a Bluetooth keyboard I own act as the head of an AI coding agent running on a remote machine — the agent's responses appear on the keyboard's small e-ink display.
Command-line tool for Baserow (an open-source Airtable alternative). Every command is generated automatically from Baserow's API spec, so the CLI stays in sync as Baserow evolves.
Command-line tool for n8n (an open-source workflow automation platform — Zapier you can self-host). Generated from n8n's API spec, same pattern as baserow-cli.
Visit a URL, your home server wakes up, you get redirected to it. The piece making that work is a tiny always-on network gadget acting as a wake-up gate.
An interactive map of who makes the memory chips in your computer — the foundries, the assembly plants, the customers — built from publicly available trade data.
A configurable status bar for Claude Code (an AI coding tool that lives in the terminal) — shows current model, git branch, token usage, and live API cost as you work.
Code from a 2025 paper of mine. Lets an old optimisation algorithm discover its own movement rules through evolution, instead of using the textbook formula written in 1995.
An earlier framework (2023) that built neural networks by evolving them — combining biological-style mutation with conventional gradient training. Codebase behind my first Q1 journal paper.