A high-stakes, AI-driven coordination system designed to transform scattered, unstructured NGO data into real-time, trustworthy volunteer deployment decisions for social impact.
- Dual AI Model Architecture:
- Gemini 2.0 Flash: Handles all multimodal (PDF, Audio, Images) and visual predictive tasks.
- Groq Llama 3.3 70B: Powers speed-critical text reasoning and volunteer matching.
- Emotion-Triggered Escalation: Gemini detects hidden distress or panic in messages/voice notes to escalate priorities that keywords miss.
- 72-Hour Crisis Prediction: Predicts regional risk scores based on 90-day history and trends, firing pre-alerts to volunteers before a formal request is even made.
- Offline SMS Fallback: A Twilio-based system that works in rural India without internet, queuing reports until connectivity returns.
- Self-Improving Feedback Loop: NGO/Volunteer ratings are stored to re-analyze low-rated matches and predictions.
- Backend: FastAPI (Python 3.11) with Motor (Async MongoDB).
- Frontend: Next.js 14 (App Router) with Tailwind CSS & Lucide Icons.
- Database: MongoDB Atlas (Free Tier).
- Geospatial: Leaflet Heatmap with Google Maps Geocoding API.
- Cache: Redis for Groq response caching.
- Security: Google Cloud Secret Manager for all API keys.
- Cloud: Deployed on Google Cloud Run (
asia-south1).
├── backend/
│ ├── main.py # FastAPI entry & Middleware
│ ├── api/ # Intelligence, Ingest, Prediction routers
│ ├── utils/ # AI clients, Secrets, Caching, Geocoding
│ ├── db.py # MongoDB configuration
│ └── schemas.py # Pydantic models
├── frontend/
│ ├── src/app/ # Dynamic Dashboard, Dispatch, Alerts
│ └── src/components/# Heatmap, Needs Queue, Risk Charts
└── Dockerfile # For Cloud Run deployment
The following keys must be stored in Google Cloud Secret Manager:
MONGODB_URIGEMINI_API_KEYGROQ_API_KEYTWILIO_ACCOUNT_SID/TWILIO_AUTH_TOKENGOOGLE_MAPS_API_KEY
To seed the platform with demo data for judges:
python -m backend.utils.demo_dataBuilt for the Google Advanced Agentic Coding Hackathon.