Whiff: A Learning Lab in Your Browser

Whiff puts a small fly world and a live brain inspector in your browser. Each fly has smell-learning circuitry derived from the FlyWire fruit-fly connectome. You can pair a scent with food, change the room and watch what it learns.

The part I enjoyed building was making the model readable. The world shows the behaviour, and the inspector shows the activity behind it. You can move from watching a fly to exploring how a learning rule works.

I built the interface in TypeScript with Three.js and WebGL. Simulation work runs in Web Workers so the controls stay responsive. Python handles the one-time preparation of the connectivity data.

The teaching section uses interactive lights and knobs to explain supervised, unsupervised and reward-based learning. It is working math you can change, with visible state and immediate feedback.

The connectivity comes from FlyWire. The smell model uses simplified rate neurons, the movement controller is artificial, and the room is a simulation. This is an educational software project, not a claim to reproduce a living animal. The pixel-art cover is an illustration. The other images are actual screenshots from the running app. Scientific data and research are credited in the project write-up.

I like building interfaces for complicated systems, especially when people need to understand what the software is doing. Whiff brings together browser graphics, simulation and interactive explanations.