Project SAFE – AI Disaster Preparedness Assistant

Inspiration

I wanted a tool that works when people are stressed and time is short. Alerts exist, but they don’t translate into what to do right now. Project SAFE turns warnings and images into calm, concrete actions.

What it does

Project SAFE combines real‑time NOAA weather alerts with Gemini 3 to deliver personalized preparedness checklists, home safety analysis, and damage assessments from photos or documents. Users can generate actionable guidance, then share or export it.

How we built it

I built the backend in FastAPI with PostgreSQL, integrated Gemini 3 (Flash Preview for text + Pro Image Preview for vision), and NOAA’s public alerts API. The frontend is React + Vite with a clean, accessible UI. I added sharing and PDF export for checklists.

Challenges we ran into

Getting OAuth, deployment, and CORS correct across Render and Vercel took time. Gemini quota limits also required careful rate limiting and fallback handling to keep the demo stable.

Accomplishments that we're proud of

• Real‑time alerts + AI reasoning in one flow
• Multimodal Gemini 3 integration (text + image)
• A demo that works without login for judges

What we learned

Real‑world AI apps need strong UX, not just strong models. The most important part is translating model output into calm, actionable guidance under stress.

What's next for Project SAFE – AI Disaster Preparedness Assistant

Add offline mode, local emergency plans, multilingual support, and deeper integrations with local emergency services for verified guidance.

Built With

• 3
• fastapi
• gemini
• noaa
• postgresql
• react+vite
• render
• sqlalchemy+alembic
• vercel