HydroShield AI

AI-powered flash flood monitoring agent delivering real-time telemetry alerts and automated safe-evacuation routing.

Comment

Inspiration

On August 26, 2026, a massive glacier and rock collapse occurred at Langtang Lirung in Langtang National Park, Nepal. The resulting debris avalanche triggered a sudden 30-foot surge of water, mud, and ice down the Bhote Koshi, Lende, and Trishuli river corridors across Rasuwa, Nuwakot, and Dhading districts.

More than 900 workers at local hydropower plants and hundreds of residents in valley settlements were caught unprepared. The primary failure wasn't just the sheer volume of water, but the absence of immediate, actionable warning before low-altitude valley passes were cut off.

What it does

HydroShield AI is an autonomous flash flood monitoring and safe evacuation agent.

  • Real-Time Telemetry Polling: Asynchronously ingests satellite precipitation rate (mm/h), topsoil moisture saturation (%), and river stage height (m) telemetry every 20 seconds.
  • Plain-English Hazard Translation: Translates complex environmental metrics into direct, actionable instructions ("River Overflowing. Trishuli Valley Pass Closed. Proceed to Himalayan High Ridge Plateau").
  • Spatial Route Guidance: Calculates high-ground evacuation routes around submerged low-water river crossings and deep-links directly into the official Google Maps app for 1-tap turn-by-turn driving navigation.

How we built it

  • Agent Architecture: Designed around the Google Gemini Agent Development Kit (ADK) Observe-Reason-Act (ReAct) loop. The agent evaluates environmental deltas, classifies risk levels into 4 tiers (0 to 3), and executes tools autonomously.
  • Mapping & Spatial Routing: Integrated the Google Maps Directions API to construct driving vectors from low valley origins (28.2150°N, 85.3850°E) to high-ground shelters (28.2580°N, 85.4200°E).
  • Interactive Visualization: Built a Single Page Application (SPA) with a custom HTML5 canvas spatial radar rendering river channels, sweeping radar scans, flood hazard polygons, and evacuation paths.
  • Backend & Telemetry: Lightweight Python 3 HTTP server (server.py) serving static assets with sub-second response times, connected to Open-Meteo REST satellite endpoints.

Challenges we ran into

  1. Resolving Route Coordinate Ambiguity: Initial mock origin and destination coordinates were too close, causing the route vector on Google Maps to look unclear. We restructured the spatial dataset so the origin is explicitly pinned to the low river valley floor (Trishuli Pass) and the destination is pinned to a high-elevation plateau shelter (Langtang Camp at 2,800m), rendering a distinct, unambiguous driving vector.
  2. Panic-Proof Human Interface: Environmental sensor numbers (like 88.5 mm/h rainfall or 7.50m river height) cause confusion during extreme panic. We built a threat classifier inside the Gemini loop that converts metrics into direct plain-English guidance.

Accomplishments that we're proud of

  • Successfully built a working end-to-end prototype combining autonomous AI agent reasoning with real-time satellite APIs and live Google Maps navigation.
  • Created a high-resolution, zero-dependency HTML5 canvas topographic map that updates dynamically as flood risk levels escalate.
  • Demonstrated a practical, life-saving application inspired by real-world disaster events in Nepal.

What we learned

  • How to structure autonomous agentic loops using Google Gemini ADK principles to handle real-time tool execution.
  • Techniques for handling REST telemetry streams and formatting spatial deep-links for mobile map navigation.
  • The importance of designing disaster response interfaces that prioritize immediate human clarity over raw data density.

What's next for HydroShield AI

  • Google Maps Traffic & Road Hazard API Integration: Ingesting live traffic congestion data and municipal road closure feeds to automatically reroute evacuees around active traffic jams and landslides.
  • IoT Acoustic River Sensor Deployment: Connecting LoRaWAN acoustic velocity sensors placed along mountain river basins for sub-second debris wave detection.
  • Physical Barrier Gate Automation: Interfacing with municipal traffic infrastructure to lower physical barrier arms automatically at submerged river crossings.

Project Repository Link:

https://github.com/DeviKamakshi/Projects

Built With

  • google-gemini-agent-adk
  • google-maps-directions-api
  • javascript
  • open-meteo-rest-api
  • python
Share this project:

Updates