AWS Builder Center

KAVALAN: Building an AI-Powered Flood Disaster Intelligence System for Chennai

How I built a ward-level flood prediction and emergency response PWA using hybrid ML forecasting, bilingual AI alerts, and real-time evacuation tracking for Chennai's flood-prone communities.

The Problem

Chennai floods almost every monsoon season, but most residents get warnings too late — or not at ward-level granularity that actually helps them act. I built KAVALAN to close that gap: a disaster intelligence system that predicts flood risk at the neighborhood level and gets alerts to people in the language they understand, before water reaches their door.

What KAVALAN Does

KAVALAN is a Progressive Web Application that combines machine learning forecasting with real-world communication tools:
  • 72-hour hybrid ML flood forecast — predicts risk windows so residents and authorities have real lead time, not just same-day warnings.
  • Bilingual AI alerts (Tamil & English) — sent via Twilio SMS, so alerts reach people regardless of smartphone or app access.
  • Community evacuation tracker — a live view of who has evacuated and who still needs help, useful for local volunteers and relief coordination.
  • Ward-level granularity — instead of city-wide alerts, KAVALAN narrows predictions down to the ward level, which is where flood impact actually varies most.

Tech Stack

  • Frontend: React.js (PWA, so it works offline and installs like a native app)
  • Backend: FastAPI for serving ML predictions and handling alert logic
  • Database/Realtime: Firebase for live evacuation tracking
  • Maps: Leaflet.js for ward-level visualization
  • Alerts: Twilio for SMS delivery

Why a PWA?

Many of the people who need flood alerts most don't have reliable app-store access or high-end phones. A PWA meant KAVALAN could be installed directly from the browser, work with poor connectivity, and stay lightweight — without sacrificing the interactive map and real-time features.

What I Learned

Building KAVALAN taught me how different "AI for social good" is from a typical ML project. The hard part wasn't just training a forecasting model — it was making the output usable: getting a prediction to translate into a message a non-technical person could read and act on in seconds, in their own language, over SMS with zero setup required.

What's Next

I'm working on expanding KAVALAN's forecasting model with more historical rainfall and elevation data, and exploring integration with local government alert systems so ward-level predictions can plug into existing emergency response workflows.


Built as part of my ongoing work in AI/ML and disaster-response tech. Would love feedback from anyone working on similar civic-tech or disaster-intelligence projects!
Any opinions in this article are those of the individual author and may not reflect the opinions of AWS.
Enjoyed reading this content? Let the author know!

Your likes, comments, shares, and saves help creators reach more builders.

Loading recommendations

Loading article