Hackathon · Jun 2026
ReVival
Seven agents that grade, price and resell returned products instead of sending them to landfill.
- EventAmazon HackOn, Season 6
- ThemeSecond-life commerce
- TeamWith Arush Sharma
Problem
Returned items are often written off. The brief asked for AI that decides whether an item is resold, refurbished, donated or recycled, grades its condition from photos and finds it a buyer.
What we built
- Grading: a vision model reads photos or video and returns a condition grade with evidence.
- Routing: each item goes to the outcome with the best value plus carbon saved.
- Pricing from grade and local demand, with a discount for nearby buyers.
- Matching: regional buyers are shortlisted, then ranked by their risk of returning the item.
- Green credits and a trust passport that records an item's history.
- FastAPI on AWS Bedrock, DynamoDB and S3, with a Next.js front end.
Design choices
- Rules decide first and the language model reranks, so every decision can be explained.
- Grades are cached by image similarity, so lookalike items skip a second vision call.
Result
I built the demand side: buyer matching, pricing, the trust passport, green credits, and the buyer and operations screens. Arush built grading, routing and the AWS setup.