SmartBin – Crowd-Mapped Public Dustbin Finder
A crowd-mapped directory of public dustbins in Indian cities, with AI-verified photo submissions and an admin dashboard. Submitted to the WeMakeDevs × AWS Environmental Hacks.
Key Highlights & Achievements
Led a team of 3 to build and submit SmartBin to the WeMakeDevs × AWS Environmental Hacks (Waste & Energy track), and owned the backend: an Express/TypeScript REST API (/v1/*) on AWS EC2 behind API Gateway, backed by PostgreSQL on Amazon RDS with Drizzle ORM and Zod request validation.
Implemented authentication with Amazon Cognito (JWT verification via aws-jwt-verify) and a direct-to-S3 photo pipeline using short-lived presigned URLs, so images never pass through the API server.
Designed a nine-layer anti-fake defense enforced on the API: live-camera-only uploads, file-type and byte-level image validation, photo ownership checks, a 15 m duplicate radius, monthly per-user limits, and missing-bin reports only within 20 m. Passes 15/15 automated anti-fake checks.
Added an AI photo check with the OpenAI vision API that separates real public bins from fakes, with a provider switch to Amazon Bedrock. On a 20-photo hand-labelled evaluation it blocked 12/12 fakes and accepted 6/6 clear bins (95% accuracy, ~1.2 s per photo).
Deployed to AWS with secrets in Systems Manager Parameter Store and an IAM role instead of access keys, plus OpenStreetMap fallbacks if Amazon Location Service is unavailable.