Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for PayPilot
PayPilot — AI Proposes. Security Decides. Payments Execute.
Inspiration
AI agents are increasingly capable of making decisions and initiating purchases. However, giving AI unrestricted financial authority introduces risks such as unauthorized spending, policy violations, and lack of accountability.
We built PayPilot to solve this problem by separating AI intelligence from financial authorization. Our goal is to make AI-powered purchasing safer, more transparent, and accountable.
What it does
PayPilot is an AI-powered purchasing governance platform that allows AI to recommend purchases while enforcing deterministic security policies before any payment can proceed.
Key features include:
- AI-powered recommendations: Google Gemini analyzes purchase requests and provides contextual recommendations.
- Deterministic security: Server-side policies independently evaluate spending limits and authorization requirements.
- Human-in-the-loop approval: Transactions requiring additional authorization are sent for human review.
- PayPal Sandbox integration: Authorized purchases can proceed through simulated payment processing.
- Audit trail: Transaction decisions and security events are recorded for traceability.
- Governance dashboard: Users can review transactions, policies, security events, and audit information.
How it works
- A user submits a purchase request.
- The AI advisory agent analyzes the request and provides a recommendation.
- The server-side security engine independently evaluates the transaction against deterministic policies.
- Unauthorized transactions are blocked, while transactions requiring approval are routed to a human.
- Authorized purchases proceed through PayPal Sandbox.
- Transaction outcomes and decisions are recorded in the audit trail.
The central principle: AI recommendations never override security authorization.
How we built it
We built PayPilot using React, TypeScript, and Vite for the frontend, with Node.js, Express, and TypeScript powering the backend.
Google Gemini provides AI-powered purchasing recommendations. PayPal Sandbox supports simulated payment execution, while SQLite stores application data.
The most important architectural decision was separating the AI advisory layer from the authorization layer. AI can recommend an action, but only deterministic server-side security rules can authorize it.
Challenges we faced
One of our
Built With
- ai
- ai-agents
- cybersecurity
- express.js
- fintech
- google-gemini
- node.js
- paypal
- paypal-sandbox
- react
- rest-api
- sqlite
- typescript