# PayProof

- **Event:** [Agents in Context Toronto: a MongoDB Hackathon](https://cerebralvalley.ai/e/mongodb-toronto-hackathon)
- **When:** Sat, Nov 14 at 9:00 AM – 9:00 PM (EST)
- **Where:** Toronto, Canada
- **Team:** [Naoya Yamashita](https://cerebralvalley.ai/u/conao3)
- **GitHub:** https://github.com/conao3/ts-payproof
- **Demo video:** https://www.youtube.com/watch?v=QneL7pynr7k
- **Gallery:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/1

PayProof helps a small Canadian construction business check a supplier's bank-account change before paying an invoice. The Canadian Anti-Fraud Centre names construction and contracting businesses among common targets of payment-redirection fraud and recommends checking bank changes through trusted contacts. PayProof makes that check a required step before an agent can execute a payment.

A real LLM extracts the invoice and calls restricted tools. The server holds a changed payment destination until a separate human approver records verification through the supplier's saved contact. Approval is tied to the exact invoice revision, amount, currency, destination, supplier revision, and policy revision. Changing the amount or destination invalidates that approval.

The demo uses synthetic business data, an organizer-provided MongoDB Atlas cluster, and a separate HTTP service that records simulated payments. It shows the bank-change hold, human verification and approval, rejection of an old approval after an amendment, and a successful simulated payment with a matching receipt and decision trail. An independent HTTP check also verified that concurrent dispatches produced one ledger entry.

A read-only voice panel uses ElevenLabs for speech recognition and synthesis. The existing LLM explains the selected invoice using the signed-in user's authorized Atlas snapshot and returns supporting records on screen. Voice cannot approve, pay, or change an account. An authenticated HTTP test used a synthetic spoken question and verified cited records without changing the invoice or ledger.

The application and its workflow were built during today's hackathon. OpenAI provides the model; the repository lists the framework and library dependencies. Payments are simulated, and supplier confirmation is a synthetic demonstration.

Problem source: https://antifraudcentre-centreantifraude.ca/news-nouvelles/2026/2026-05-13-eng.htm
Website: https://conao3.github.io/ts-payproof/
Pitch: https://conao3.github.io/ts-payproof/slides/

## More from Agents in Context Toronto: a MongoDB Hackathon

- [Bodega](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/2)
- [One Lesson](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/3)
- [Solo Spades](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/4)
- [Travelwanders](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/96)
- [Stephen/Soham](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/97)
- [Day One](https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/98)

---

Markdown version of https://cerebralvalley.ai/e/mongodb-toronto-hackathon/hackathon/gallery/1. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt