DrainSage - Lagos Flood Safe

AI vision that sees blocked drains, reasons flood risk with Bedrock Claude 3.5, acts via SMS alert. Prevents Lagos floods. Built COOL on Graviton4, S3, Lambda, SNS.

Comment

Inspiration

What## Inspiration

Lagos floods every rainy season - not because we lack drains, but because drains are blocked with plastic waste. In 2024 alone, Lagos recorded 200+ flood incidents. As a Lagosian, I have seen streets turn to rivers. I wanted to build a vision system that SEES the blockage, REASONS the flood risk, and ACTS before disaster strikes.

What it does

DrainSage is a 7-step AI vision system:

  1. Community worker takes a photo of a drain with phone
  2. Frontend (GitHub Pages + Netlify) uploads image
  3. AWS Lambda on Graviton4 ARM64 runs OpenCV 5 + UNet 3.2M model to SEE blockage %
  4. Image stored in Amazon S3 Data Lake
  5. Amazon Bedrock Claude 3.5 Sonnet REASONS: Is it high risk? What action needed?
  6. Amazon SNS sends SMS alert to LAWMA team
  7. Outcome: Flood risk mitigated, community safe

Live demo: Upload any drain photo, get instant blockage detection + AI reasoning.

How we built it

  • Frontend: HTML/JS on GitHub Pages (https://josephine2016.github.io/Drainsage-lagos-flood-safe/) + Netlify backup
  • Backend: Serverless Lambda Function URL on AWS Graviton4 - 40% cheaper, low carbon, COOL optimized
  • Vision Model: OpenCV 5 + UNet 3.2M parameters for segmentation
  • Storage: Amazon S3 for image lake
  • Reasoning: Amazon Bedrock Claude 3.5 Sonnet
  • Action: Amazon SNS for SMS alerts
  • Theme: Build vision systems that see, reason, act

Architecture: 7-box flow (Community Phone -> Frontend -> Lambda Graviton4 -> S3 -> Bedrock -> SNS -> LAWMA Safe)

Challenges we ran into

  • GitHub Pages deployment with index.html nested in folder
  • Lambda cold start with OpenCV layer - solved with ARM64 Graviton optimization
  • Making UNet model lightweight enough for Lambda 512MB
  • Converting webp images to png for Devpost gallery

Accomplishments that we're proud of

  • 3 LIVE deployments: GitHub Pages, Netlify, and AWS Lambda Function URL
  • 7-step architecture working end-to-end
  • Graviton4 COOL optimization - 40% cost saving, low carbon
  • Real impact for LAWMA and Lagos flood prevention

What we learned

  • How to deploy serverless computer vision on AWS Lambda + Graviton4
  • How to chain S3 + Lambda + Bedrock + SNS for see-reason-act pattern
  • How OpenCV UNet can run in serverless with optimization
  • Importance of community-driven AI for African cities

What's next for DrainSage

  • Add IoT water-level sensors
  • Add Yoruba/Pidgin SMS alerts
  • Partner with Lagos State LAWMA for pilot in 3 flood-prone LGAs
  • Add map dashboard of high-risk drains

it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for DrainSage - Lagos Flood Safe

Built With

Share this project:

Updates

Submission history