Inspiration
Online deals run on trust. Buyers worry about paying for work that doesn't match the brief, and sellers worry about delivering and not being paid. Disputes are slow and costly for everyone, including the payment platform. We wanted to see if PayPal's payment holds plus an AI referee could settle this automatically.
What it does
ProofPay is an AI-judged escrow built on PayPal.
- The buyer posts a brief and an amount. PayPal creates an order with the authorize intent, so the money is held, not charged.
- The seller submits the delivery.
- A Gemini AI referee turns the brief into a checklist, scores each item pass or fail with a reason, and approves or rejects.
- If approved, PayPal captures the payment. If rejected, PayPal voids the hold and no money moves.
How we built it
PayPal Orders and Payments APIs (sandbox): authorize, capture and void. Google Gemini returns a structured JSON verdict. The backend is Python with FastAPI, and the frontend is a single HTML and JavaScript page that updates live while waiting for the buyer's approval. The AI decides before any money moves.
Challenges we ran into
A seller account set up in India could not accept USD, so we moved to US sandbox accounts.
- AI timeouts and high-demand errors, handled with retries and model fallback.
- Keeping money safe: if the judge fails, the payment stays on hold.
Accomplishments that we're proud of
What we learned
How PayPal's authorize, capture and void flow works, and how to put an explainable AI decision in front of a payment action.
What's next for ProofPay
Milestone payments with partial capture, a dispute evidence report, PayPal webhooks, and a human override for borderline cases.
