Inspiration

Small sellers on PayPal lose money to disputes, not just from fraud but from missing paperwork. Building a response by hand takes ~40 minutes, deadlines are tight, and a dispute rate above ~1% can put funds on hold. We wanted sellers to be ready before a dispute happens.

What it does

ProofPilot is a seller console that protects PayPal sellers proactively:

  • Evidence at checkout: an AI agent builds an evidence pack the moment an order is placed.
  • Risk scoring: every order gets a risk score so sellers know which ones need attention.
  • Evidence boosting: adding tracking lifts the pack's strength and registers it with PayPal.
  • Hold early warning: monitors dispute rate against PayPal's ~1% comfort level.
  • Auto-assembled disputes: when a buyer opens a dispute, the case arrives with evidence checked and a cited response drafted, ready for the seller to review and submit.

How we built it

  • Frontend: React + Vite (storefront and seller console)
  • Backend: Node.js server with PayPal REST APIs (sandbox)
  • AI agent: Google Gemini 2.5 Flash for evidence analysis and cited response drafting
  • Storage: MongoDB

Challenges we ran into

  • Making AI-drafted responses cited and grounded in real order evidence, not hallucinated
  • Simulating the full checkout-to-dispute lifecycle without live sandbox credentials
  • Designing evidence-strength scoring that sellers can understand at a glance

Accomplishments that we're proud of

Cutting case-building from ~40 minutes by hand to an instantly drafted, cited response, and shifting dispute handling from reactive to proactive.

What we learned

Evidence quality at the moment of purchase matters more than scrambling after a dispute.

What's next for ProofPilot

Live PayPal sandbox/production integration, more evidence sources (shipping carriers, buyer communication), and multi-seller support.

Built With

Share this project:

Updates

Submission history