Sentinel AI
AI that predicts floods, verifies citizen reports in under 3 seconds, and auto-dispatches volunteers across the full disaster lifecycle.
Built at NovaStark Hacks 2026
Team TECH MAX: varun, rishanth patkar, Manasvi Gudey
Built with Python, Flask, PostgreSQL, PostGIS, SQLite, Docker, APScheduler, Twilio, OpenAI, NVIDIA Nemotron, Qwen3, LangGraph, Uber H3, Leaflet.js, MapLibre GL, Open-Meteo, RainViewer, NASA GIBS, Chart.js, Bootstrap 5, PWA (Service Worker), Render
About the project
PROBLEM: Disaster response in India fails before it starts. Agencies work in silos, field reports take hours to verify and get duplicated, mobile networks collapse mid-disaster, alerts miss non-English communities, and survivors' needs never reach donors. India lost $79.5B to climate disasters in 1998-2017 (UNDRR), with 100Cr+ people in disaster-prone zones.
SOLUTION: Sentinel AI is a predictive risk intelligence and coordinated response system, flood-first and extensible to cyclone, surge and fire. It covers the full lifecycle on one shared Uber H3 risk grid.
BEFORE: A live Digital Twin ingests rainfall radar, satellite, river and government data, rescanned every 5 minutes, and scores Risk = Hazard x Vulnerability per cell. A LangGraph Prediction Agent turns forecasts into hazard windows (start, peak, ease), not just "how bad".
DURING: Citizens report via online form, one-tap Voice SOS, or an AI Calling Agent that works on any phone with no mobile data, in 6 languages (English, Hindi, Telugu, Tamil, Malayalam, Kannada). Every report gets a 4-signal AI score (image, heatmap corroboration, user quality, climate alignment; 25% each), with screen-recapture and duplicate detection. >=85% goes to priority dispatch, 50-84% to official review, <50% is held for corroboration. Geo-fenced volunteers are notified via WhatsApp (Twilio) in under 3 seconds (tested), and nothing closes without GPS + photo proof.
AFTER: LifeLine P2P matches survivors' needs with donors' supplies automatically.
HOW IT WORKS: OpenAI leads verification with NVIDIA Nemotron as fallback so scoring never fails. Offline-first PWA syncs on reconnect, with voice call + SMS fallback. Command dashboards (supervisor, analyst, municipal commissioner) give a ranked queue, approve/reject, and a timestamped, CSV-mirrored audit trail.
IMPACT: Stops duplicate deployments, cuts verification from hours to seconds, warns only people actually at risk, and keeps every action auditable. 26+ regions monitored, Digital Twin live in 2 cities, no dedicated hardware needed. Pilot cost is about Rs 8-10K per city per month. Roadmap: Hyderabad (GHMC), then Bengaluru, then a coastal district, via MoU with SDMAs and EOC integration. DPDP Act-compliant self-hosted processing is planned before scale.