Inspiration

Splitting costs after a trip or dinner often means scrolling through a long group chat, figuring out who paid, and doing the math by hand. We wanted to make settling up as easy as sharing the chat.

What it does

Split turns a pasted group chat into an expense breakdown and calculates who owes whom. Users can choose a country and currency, review the suggested repayments, and open PayPal. My links to square up. The links do not automatically send payments.

How we built it

We built the app with Next.js and React. The parsing API uses the Vercel AI SDK with Gemini and a Zod schema to extract structured expenses. The app then calculates balances in code, rather than asking AI to do the settlement math. We also added currency-aware formatting, regional sample names, and an installable PWA experience for mobile.

What we learned

We learned that extracting expenses is only part of the challenge: the settlement math needs to be predictable and easy to verify. Currency precision matters too—some currencies don’t use fractional units, so repayments must be rounded without leaving the group’s ledger unbalanced.

Challenges

Messy chat messages can be ambiguous, and people may not participate in every expense. We addressed this by letting users review the breakdown and accounting for explicitly excluded participants. The prototype assumes each chat uses one currency; it doesn’t convert mixed-currency expenses. PayPal. My usernames and amounts should also be checked before paying.

Built With

  • google-gemini
  • next.js
  • paypal
  • progressive-web-app-(pwa)
  • react
  • vercel
  • vercel-ai-sdk
  • ypescript
  • zod
Share this project:

Updates

Submission history