Inspiration

I'm a solo freelancer in India, and most of my clients are abroad. Every deal lives in a chat: "logo + brand guide for $450, 50% advance, rest by Friday." Turning that into an invoice and chasing late payers is the part every freelancer puts off, and payments slip by weeks. PayPal has invoices, checkout and reminders; AI can bridge how freelancers talk to clients and a structured invoice.

What it does

InvoicePilot is an AI billing agent. Paste a WhatsApp chat or email (English, Hindi or Hinglish) and AI extracts client, line items, currency, due date, terms and advances already paid. One click creates and sends a real PayPal invoice (Invoicing v2 API) or a checkout link (Orders v2 + JS SDK Smart Buttons). An agent chat answers "Who hasn't paid?" from live PayPal data and drafts reminders that PayPal sends. Nothing goes out until the freelancer approves.

How we built it

Node.js + Express calling PayPal REST APIs directly (OAuth, Invoicing v2, Orders v2, JS SDK). Google Gemini via an OpenAI-compatible client does JSON extraction, a tool-calling agent and reminder writing, with fallback models and a rule-based Hinglish parser. Demo mode runs with zero keys. Deployed on Render (render.yaml + Dockerfile), with end-to-end tests.

Challenges we ran into

Messy mixed-language chats ("k" amounts, split prices, advances), PayPal's exact item_total math (done in minor units), agent safety (agent prepares, human executes), and AI overload (automatic fallbacks).

Accomplishments and what we learned

A full loop, chat to invoice to paid to reminder, working on the real PayPal sandbox, and it understands Hinglish. Showing the AI's assumptions builds trust, and human-in-the-loop makes people comfortable letting an agent near their money.

What's next

PayPal webhooks for instant paid alerts, a WhatsApp/Telegram bot, recurring retainers via PayPal Subscriptions, GST-ready invoices, and packaging for the PayPal Agent Toolkit / MCP.

Built With

Share this project:

Updates

Submission history