Inspiration

Payment fraud costs businesses billions annually, yet most small merchants lack tools to detect suspicious transactions before they happen. We built PayGuard to make fraud detection as simple as typing a sentence.

What it does

PayGuard is an AI payment agent that lets you send money using natural language — "Send $50 to [email protected]." Before executing, it analyzes the transaction for fraud risk, scoring factors like amount anomalies, recipient reputation, and velocity checks. Low-risk transactions auto-approve; high-risk ones get flagged with explanations.

How we built it

  • Backend: FastAPI with OpenRouter (free models) for intent parsing and fraud analysis
  • Payments: PayPal Sandbox API for real transaction execution
  • Frontend: Vanilla HTML/JS with a chat-style interface
  • Deployment: Render for hosting

Challenges

  • Integrating PayPal's sandbox API with natural language intent parsing
  • Designing a fraud scoring system that's explainable, not just a black box
  • Keeping everything free-tier: OpenRouter free models, Render free tier, PayPal sandbox

What's next

  • Webhook notifications for transaction alerts
  • Multi-currency support
  • ML-based anomaly detection trained on real transaction patterns

Built With

Share this project:

Updates

Submission history