04 /CLOUD / CIVIC TECH
HYDROSENSE.
Tells Delhi drivers which underpasses are likely to flood, using rain forecasts, past floods, photos from the public and a water sensor.
How it works
Know the water before you drive into it.
Flood risk where a wrong answer is dangerous: missing forecast data resolves to "unknown", never "safe". Serverless on Lambda and DynamoDB for spiky monsoon traffic, presigned S3 uploads with size and type limits, Bedrock for photo checks, and a leave-one-storm-out backtest instead of a guess.
The hard parts
- Showing "unknown" instead of a false "safe" when forecast data is missing.
- Taking photo uploads from the public safely, with size and type limits and short retention.
Results
- Risk tiers with a "why this tier?" explanation for 16 underpass hotspots.
- Evaluated against historical flooding with a leave-one-storm-out method.
- 57 backend tests.
Stack
Reach the end of the file to complete this operation
