Jarvis

Jarvis is my senior capstone, built with four teammates. The plan is a model that learns from past disasters and estimates what the next one would cost a city in people and money, so officials and first responders can plan before it hits. We're starting with wildfire. My teammates are on the research, the data sources and the model; I built the parts that run today.

The map is Leaflet and plain JavaScript, served by a small FastAPI app. It pulls 10 layers live from 7 public sources: NIFC fire perimeters colored by containment, NASA VIIRS satellite heat detections, NWS red flag warnings, USGS fire and EMS stations, lakes and reservoirs as water sources, USFS wildfire hazard, building and population density, LANDFIRE fuel types, and the U.S. Drought Monitor. Click anywhere and it reports the fuel type, current weather and wind, and drought level there. Each of those lookups fails on its own, so one slow source doesn't blank the rest.

The other piece is the team's data pipeline. Teammates drop CSV or Excel files or add source links in the repo on GitHub, and a GitHub Actions job loads them into Supabase Postgres, rebuilding each table so a re-upload never duplicates rows. Nobody needs anything installed to add data. It has loaded 356 EM-DAT wildfire records as seed data for the planned model.

The prediction model, which is the point of the whole project, isn't built. The other hazards in the plan (earthquakes, hurricanes, tornadoes and tsunamis) are placeholders, and nothing is hosted yet, so the map only runs locally.