STORMFORGE — Critical Access Intelligence

Inspiration

After a hurricane, knowing where flooding happened is one thing, but figuring out which neighborhoods are cut off and where to send limited emergency resources is a completely different problem. I wanted to see how I could use real geographic data, graph algorithms, and AI to help make those decisions.

What it does

STORMFORGE is a hurricane-response simulator focused on St. Augustine, Florida. You can choose a hurricane category, see which roads are affected by flooding, and identify neighborhoods that have lost access to emergency services.

You can also deploy simulated resources like debris crews, EMS teams, and rescue units to see how much they actually help. I integrated Google's Gemma AI to recommend resource allocations and compare them against your own decisions.

How I built it

I used Python, FastAPI, and NetworkX for the backend, with React, TypeScript, and MapLibre for the frontend. The project uses NOAA SLOSH storm-surge data and OpenStreetMap roads to build a road network and calculate accessibility.

Gemma handles the AI recommendations, but the backend independently validates the allocations and calculates their effects rather than relying on AI-generated numbers.

Challenges and what I learned

The hardest part was getting the geographic data and road-network calculations to work together. At one point, my hospital-access results were misleading because the road network didn't properly account for nearby emergency departments. Fixing that taught me how important it is to validate the model, not just make sure the code runs.

I also learned a lot about integrating AI with traditional algorithms and making complicated technical information understandable through a simple interface.

What's next

I'd like to expand STORMFORGE to more locations, improve the accuracy of emergency routing, and eventually incorporate travel times and additional infrastructure data.

For now, it's an educational simulator rather than a real emergency dispatch system, but it shows how geographic data and AI can be combined to explore disaster-response decisions.

Built With

• artificial-intelligence
• fastapi
• geospatial-analysis
• google-gemma
• graph-algorithms
• javascript
• maplibre
• networkx
• noaa
• openstreetmap
• python
• react
• rest-api
• slosh
• typescript