Agentic engineer London, UK
Building tools
for AI agents.
I build tools that let AI agents work with desktop apps, graphics, video, and creative workflows—with results people can inspect and revise.
Systems engineering at SimCorp. Previously Eigen Technologies and JPMorgan. MEng Computer Science, UCL.
01 / Selected work
Featured / Desktop tools for agents
Gives an AI agent its own background Windows desktop so it can use and test applications while you keep using your PC. Includes a live viewer, screenshots, accessibility controls, and an optional virtual gamepad.
Separate input sessions, running as the same user with access to the same files.
Graphics for agents
Local vector and raster editing for diagrams, illustrations, and image assets. Agents preview and revise artwork, keep editable sources, and export the result.
Video for agents
A local video editing engine for agents. Arrange footage, add captions and graphics, mix sound, and export—with saved revisions and undo.
Pixel art for agents
Pixel art and animation from editable JSON recipes. Agents make, preview, and revise sprites and atlases; people can edit the same recipe in a local studio.
Fabrication & agents
Arrange parts on fixed sheets for laser, CNC, plasma, waterjet, and fabric cutting. Runs locally in a browser, through the CLI, or as an MCP server.
Also: AgentUI gives coding agents forms, sliders, diffs, and live previews for decisions that need a person.
Every pixel-art illustration on this page was made with Pixelbolt.
Writing with coding agents
Start with a brief, review the story bible, then let Codex or Claude Code plan and draft LitRPG chapters. The rebuilt Python tool keeps story state, checks, and call records in plain files.
The generation pipeline works; assessing literary quality remains an open research problem.
29 September 2026 release: fast-brain-11 passes the Minus Two hairpin with the round-6 pilot stack and assisted yaw. This development flight later crashed; it is not a completed race.
Learned control & simulation
A connectome-constrained recurrent network trained to fly a drone in Liftoff. Its 30,000-neuron flight circuit uses fruit-fly connectivity, with learned parameters and a differentiable PyTorch simulator.
Published race result, 25 September 2026: fast-brain-08 completed three laps of Straw Bale in 5:17.805 with explicit pilot assistance.
Research prototype. The newer fast-brain-11 improves obstacle handling but has no completed race. Reliable unseen-course racing and freestyle flight remain unproven.
GPU simulation & artificial life
A GPU artificial-life laboratory with five worlds and 42 presets. Physarum, Particle Life, Lenia, and reaction–diffusion are joined by Symbiosis, which couples agents and chemistry. Includes saved recipes, live measurements, novelty search, and headless image and video export.
Preview of the four original worlds. Explore Symbiosis and the current presets in the repository.
Pixel art & animation
Turn JSON recipes into pixel art, sprite sheets, and animations. A local browser studio adds live editing, onion skinning, and export; CLI and MCP controls let agents use the same workflow.
Game engines
An unofficial American McGee’s Alice re-remaster in a new Rust engine, with controller support and guided setup. All 39 campaign visits are playable in an experimental Windows preview; compatible game data is required separately.
Music library tools
Save Spotify library metadata in portable formats, with an optional local companion for matching and migration to YouTube Music.
Languages & geography
Explore official languages and language families across countries, compare overlapping languages, and share the current map view.
Global trade
Compare countries, goods and services, and export markets using UN Comtrade and World Bank data.
Maps & modelling
Compare approximate travel times across London by walking, rail, bus, and car.
Language learning
Practise Mandarin pinyin typing with adaptive difficulty, pronunciation audio, and Simplified/Traditional characters.
Language learning
Practise radicals and common Chinese characters with pinyin and meaning quizzes, spaced repetition, and mastery tracking.
Generative models
An experiment in steering generative images through semantic directions, with keyboard navigation.
02 / Google Play
Three educational quiz apps covering language and history.
Find them on Google Play
LanguageLearn the Greek alphabet through symbol recognition and alphabet-order quizzes.
View on Google Play
HistoryIdentify historical armour through images, descriptions, and the periods it belongs to.
View on Google Play
HistoryPut a name to the portrait and explore the emperors and dynasties of ancient Rome.
View on Google PlayAt SimCorp, I architected and now operate a distributed calculation platform in production.
My work spans worker coordination, batch scheduling, the SQL data layer, and investigating failures across data and code. Previously, I worked at Eigen Technologies and JPMorgan, and founded an education startup.
MEng in Computer Science from UCL, specialising in AI.
I design tools around how agents work: discoverable operations, previews, saved state, and changes people can inspect and revise. I use Codex and Claude Code for implementation, debugging, and testing, while setting the architecture, reviewing changes, and verifying results.
