Inspiration
Digital payments are becoming increasingly automated, but automation also introduces risks. A payment decision should not depend only on whether a transaction can be processed. It should consider merchant trust, spending limits, transaction risk, and human authorization.
This inspired us to build PayGuard AI, a payment authorization and security control plane that puts safety and transparency at the center of automated payments.
What it does
PayGuard AI combines deterministic, rule-based orchestration, risk assessment, policy enforcement, human approval, and PayPal Sandbox checkout.
Its five specialist agents work together:
- Planner: Converts a purchase mission into constraints and steps.
- Merchant: Evaluates merchant context.
- Risk: Calculates transaction risk.
- Policy: Evaluates spending rules and recommends a decision.
- Supervisor: Combines the results and requests human authorization.
A policy recommendation is not payment authorization. Human approval is required before a PayPal order can be created.
After approval, users can continue to PayPal Sandbox checkout. PayGuard verifies the payment result before marking a transaction as completed.
How we built it
We built PayGuard AI with a Python FastAPI backend, a Next.js frontend, and PostgreSQL for transaction and application data.
The application integrates with the PayPal Developer Sandbox for controlled payment testing. Its orchestration agents use deterministic rules and existing risk and policy engines rather than independent LLMs.
We also implemented transaction monitoring, an approval center, risk analytics, policy controls, and an audit trail.
Challenges and learning
One of our main challenges was separating a policy recommendation from actual human authorization and payment execution.
We addressed this by designing a transaction lifecycle that distinguishes pending approval, human approval, PayPal checkout, and verified payment completion.
Building PayGuard AI helped us understand how payment APIs, risk assessment, policy enforcement, auditability, and human oversight can work together.
What's next
Our vision is to extend PayGuard AI with stronger risk intelligence, configurable authorization policies, improved monitoring, and more advanced AI-assisted decision support.
PayGuard AI — Secure AI Payments. Built Around Trust.
Built With
- agents
- ai
- docker
- fastapi
- llm
- next.js
- payment
- paypal
- paypal-sandbox
- postgresql
- python
- rest-api
- risk-management
- security
- typescript
