A connected
starting point.
A reproducible pipeline prepares a subgraph from the MaleCNS fruit-fly connectome: 4,096 neurons and 208,165 edges. That connectivity informs the neural architecture.
↖ ALL WORK
Bug Brain
01 / AI & FULL-STACK ENGINEERINGSeptember 2026 – Present
Bug Brain
Chess.
Neuro-inspired Chess AI
A tiny brain.
A different kind
of chess.
RECORDED FROM THE REAL APPLICATION00:32 / DEMO
4,096Neurons in the subgraph
208,165Connections between them
68Frontend & backend tests
Biology as
a starting point.
Software as the experiment.
A full-stack chess AI built around a 4,096-neuron, 208,165-edge subgraph of the MaleCNS fruit-fly connectome. A live interface brings its chess gameplay and 3D neural visualisation into the same view.
Explore how a biological connectivity graph can inform a chess system, and make the resulting software understandable through a responsive visual interface.
Recurrent neural processing, trainable policy and value interfaces, and shallow search form the decision system. A FastAPI backend connects model and game state to a React/TypeScript frontend, while WebSocket telemetry drives a Three.js neural view beside animated chess gameplay.
02 / FROM CONNECTIVITY TO A MOVE
Inside the
Inside the
decision.
01 / CONNECTIVITYDRAG TO ROTATE ↔
Conceptual system map. The actual neural visualisation is shown in the recording above.
A connected
From recurrent
state to a move.
PyTorch supports recurrent neural processing and trainable policy and value interfaces. Shallow chess search explores candidate moves; Stockfish supplies opponents and evaluation.
An ongoing engineering experiment in neuro-inspired game AI.
See the game.
Inspect the system.
FastAPI connects model and game state to a React and TypeScript frontend. WebSocket telemetry powers a real-time Three.js neural view alongside animated chess gameplay.
Replay and Elo tracking make games reviewable. Model and game state persist across sessions.
03 / ENGINEERING THE EXPERIMENT
Built to
Built to
keep exploring.
68 automated tests span the frontend and backend. Reproducible data processing and persistent state support an experiment that can be revisited and developed further.
- 4,096 neurons and 208,165 edges
- Stockfish-based opponents and evaluation
- Animated gameplay, replay and Elo tracking
- Persistent model and game state
- Reproducible data processing
- 68 automated tests across frontend and backend