PINNS TSUNAMI WARNING SYSTEM
The gap
Physics-Informed Neural Networks can model wave propagation using the actual governing equations — potentially faster and more accurate than traditional sensor-threshold systems for tsunami warnings. But this research lived in Jupyter notebooks. Nobody had built an interface where you could see seismic events on a globe, watch data streams update in real time, and understand what the model was predicting. The research was sound. The visualization didn't exist.
What I built
A 3D globe visualization using CesiumJS and Mapbox that displays live seismic data via WebSocket feeds. You spin the globe, see events as they come in, and watch wave propagation predictions rendered in real time. React/Vite frontend, backend on Railway. The goal was research-grade: accurate coordinates, proper projections, real physics. Not a demo that looks impressive but means nothing.
CesiumJS is powerful and underdocumented
CesiumJS assumes you're building Google Earth, not a research tool. Getting custom data layers to render on a 3D globe with correct projections took more effort than the actual data pipeline. WebSocket reliability was another problem — connections drop, data arrives out of order, and you need graceful degradation because this is monitoring software that researchers leave running. The PINN model integration required translating between the ML pipeline's coordinate system and geographic coordinates without introducing errors that compound over distance.
Where it is now
Deployed and live. Spin the globe at the link above and watch real seismic events appear as they're detected. Active research visualization platform at UTSA.