NeuroAtlas

Neuroscience is taught one scale at a time: a lobe diagram in one chapter, a synapse cartoon in the next, an action-potential graph somewhere else. Disease cuts across all of them at once, and that connection is exactly what static figures cannot show.

NeuroAtlas puts every scale into one navigable 3D space, from the whole brain down to tissue, neurons, synapses, and the ion channels in a single membrane, and lets you watch a disorder land on each of them.

• One typed data model for every scale. Structures, neurotransmitters and disorders live in plain TypeScript data, and each disorder declares its effects per scale (atrophy, demyelination, tangles, BBB leak, reactive astrocytes), so a new disease is data, not new scene code.
• A procedural cortex. The brain surface is generated with noise, and every vertex is assigned to a lobe and functional area (Broca, Wernicke, motor, visual cortex) using the central and Sylvian fissures as boundaries. No multi-megabyte mesh to download, and every region is clickable.
• A real action-potential model, not a looping animation. Membrane voltage is computed over time with separate modes for normal firing, persistent sodium current, and pump failure, and the channels on screen follow the same curve.
• Zustand as the single source of state, so the sidebar, info panel, search (⌘K) and five WebGL scenes stay in sync without prop drilling.
• Mobile and SEO treated as product features: a mobile guide, per-scale metadata, and generated Open Graph images.

Complex, data-heavy 3D UI, done with production discipline: a typed domain model, scene code that stays dumb, and state that stays in one place. The same skills that go into building an AI product's interface.