Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
Inspiration
Small businesses often struggle with late payments. Chasing overdue invoices takes time, creates uncertainty, and can damage customer relationships. At the same time, allowing an AI system to make financial decisions without human oversight creates serious risks.
We wanted to solve both problems.
PayPilot AI was inspired by one question: What if an AI agent could help businesses recover payments while keeping humans in control of every important financial action?
What it does
PayPilot AI is an evidence-first AI payment recovery agent designed to help small businesses manage overdue invoices.
Instead of simply generating reminders, it follows a structured recovery workflow:
- Prioritize: Identify invoices that need attention using transparent, deterministic rules.
- Investigate: Review invoice status, payment evidence, and relevant context.
- Recommend: Generate recovery suggestions and communication drafts.
- Request approval: Require explicit human approval before sensitive actions.
- Execute safely: Use controlled PayPal Sandbox payment workflows.
- Verify: Confirm payment results through PayPal rather than trusting an AI-generated response.
- Audit: Maintain an activity trail of important actions.
How we built it
We built PayPilot AI using Node.js 22 and a lightweight web application designed to work on mobile and desktop.
The application integrates with PayPal Orders API in Sandbox mode for payment workflows and uses PayPal webhook verification for payment-event handling.
Gemini provides AI-assisted explanations and communication drafts, while deterministic application logic remains responsible for financial decisions and safety checks.
We also implemented safeguards such as:
- Human approval before sensitive payment actions.
- Time-limited approvals with a 30-minute expiry.
- Revalidation of invoice details before execution.
- Blocking safeguards for disputed or already-paid invoices.
- Protection against untrusted instructions in invoice notes.
- Payment verification and duplicate-event handling.
- An audit trail for important activities.
Challenges we faced
One of our biggest challenges was balancing AI autonomy with financial safety.
An AI agent should not be able to independently decide to move money simply because a prompt tells it to. We addressed this by separating AI assistance from deterministic payment controls and requiring human approval for sensitive operations.
Another challenge was ensuring that payment status is based on verified PayPal responses rather than assumptions made by the AI.
We also focused on making the application accessible through a simple, mobile-friendly interface.
What we learned
Building PayPilot AI taught us that a useful financial AI agent needs more than a powerful language model.
It needs reliable integrations, clear authorization boundaries, evidence-based decisions, payment verification, and accountability.
We learned how to combine AI-assisted workflows with PayPal APIs, implement approval safeguards, and design an agent that supports people instead of replacing their judgment.
Impact and vision
Our goal is to make payment recovery less stressful for small businesses.
PayPilot AI demonstrates how agentic AI can help businesses act faster while preserving human control over financial decisions.
Our principle: Let AI investigate and recommend. Let humans authorize. Let PayPal verify.
What's next for PayPilot AI
Built With
- agentic
- ai
- api
- automation
- css
- design
- financial
- fintech
- gemini
- html
- javascript
- mobile-first
- node.js
- orders
- payment
- paypal
- processing
- rest
- technology
- webhooks
