Inspiration

Travel planners need to turn unstructured requests into a clear scope and a payable deposit.

What it does

Wayfare uses free local FLAN-T5 inference to summarize the brief. A planner reviews and edits the scope, exclusions, total and deposit. The customer opens a separate quote page and pays through PayPal Sandbox. The server verifies capture status, amount and currency before displaying a receipt.

How we built it

Node.js, HTML/CSS, Transformers.js, quantized FLAN-T5-base, PayPal JavaScript SDK and Orders v2 REST API. Inference runs on the local server after a first-time model download and needs no paid AI key.

Testing and challenges

The app completed a EUR60 Sandbox capture on 11 November 2026: 08H211647C761260E. No real funds were charged. Three server tests passed. A declined test card exposed the need for useful errors and retry handling. Model summaries can omit details, so planner review is essential.

What we learned

AI drafting, human approval and server-verified payment each have a distinct role. The model never controls the price or confirms a booking.

Next steps

Add planner authentication, durable storage and rate limits. The prototype uses expiring in-memory quotes.

Run it

Complete setup instructions and an open-source license are in https://github.com/fli0193429-bit/wayfare. Use your own Sandbox app and test funds.

Built With

Share this project:

Updates

Submission history