Neural Compass
BrainStorm BCI Hackathon 2026, Track 2 · solo prototype
A BCI-guidance hackathon prototype that turns simulated micro-ECoG streams into electrode-placement cues, using band-power scoring and Kalman-filtered tracking.
IllustrationHighlights
- Streams simulated micro-ECoG over WebSockets to a browser client
- FFT band-power tuning score (high-gamma up, beta suppression) with a weighted-centroid hotspot detector
- Two-axis Kalman filter for position smoothing and a SEARCHING/APPROACHING/LOCKED state machine
- Glanceable status display with a colorblind-safe Viridis palette
How it was built
For the BrainStorm BCI Hackathon (Track 2) I built a prototype that streams simulated micro-ECoG signals from a FastAPI server over WebSockets to a React client. The server turns each time window into FFT band power, computes a tuning score that rewards high-gamma activity and beta suppression, and finds a weighted-centroid hotspot. A two-axis Kalman filter smooths the tracked position, and a small state machine moves the display between searching, approaching and locked. I packaged it with Docker Compose. Limits: simulated data only; no latency, clinical or operating-room claims.