Inspiration:
"Sending money abroad is confusing. You never see the real cost, and it is easy to overspend. I wanted a helper that explains things in simple words, but never moves money on its own."
What it does:
"You type a payment in simple words. The agent understands it, checks your spending limits, compares three payment routes using a live exchange rate, and explains the result. If a payment is over your limit, it gives you choices: approve once, raise the limit, split the payment, or schedule it. You can then pay through a real PayPal Sandbox checkout, and every payment is saved in a searchable AG Grid history table."
How we built it:
"A Java backend with no framework, Google Gemini for reading the sentence and writing the explanation, the PayPal Orders API (create, approve, capture), the Frankfurter API for live rates, AG Grid for the history table, Postman for API tests, and Render for hosting. The AI never decides about money. Plain Java rules decide, so the logic can be tested."
Challenges:
"Making the PayPal approval and return flow work online, keeping the AI from saying things like 'declined' when a payment is only on hold, and handling the small free AI quota. I added answer caching and a simple fallback so the demo does not break."
Accomplishments:
"A full flow from sentence to PayPal payment, 12 passing rule tests, and a Postman collection with 6 passing checks."
What I learned:
"Keeping the AI separate from money decisions makes the system safer and easier to explain."
What's next:
"Real fee data, a database for history, user accounts with custom limits, and more currencies."
Be honest:
"Fees in the comparison are assumptions, PayPal is in Sandbox mode, and history resets when the server restarts."