Data Bounties

This project simplifies ML training by allowing companies to request images directly. They place a “bounty” for say, 10 cat photos, and then users can respond to the bounty. Upon responding, we confirm that the image is not AI, matches the prompt, and is unique. The requesting company then pays out in HBAR to a wallet created by Privy and Hedera on signup.

React + TypeScript is the frontend for browsing bounties and uploading images. Node/Express is the backend with SQLite for orchestration and state. FastAPI ML service is used for image verification. The backend receives submissions, checks for duplicates with perceptual hashing, calls the ML verifier, and if a submission passes, triggers Hedera flows to mint an NFT certificate, log the verdict, and send HBAR payout. Privy gives us fast authentication and embedded wallet UX, which made signup much smoother for non-crypto users. Hedera let us turn a successful verification into an on-chain record, payout, and collectible proof of contribution. Google Cloud helped on the ML side by powering image analysis for prompt matching and caption generation without us needing to fully self-host heavy models. One notably hacky but effective choice was keeping the ML service behind a very strict JSON contract so the backend could integrate early with a stubbed verifier before the real model pipeline was finished.