Brosplit

Problem

Group expense apps (Splitwise et al.) handle the happy path well but fall down on the edges — one currency only, leftover paise / cent disappear into rounding errors, settlement is one-sided ("I paid for you, you owe me, the end"). Brosplit handles the messy version of the problem: groups that span currencies, splits that don't divide evenly, and repayments that need the creditor's blessing.

Architecture

Next.js 15 frontend with Supabase Auth using HTTP-only cookie sessions and 7-day token-based group invitations. The split-calculation engine operates per currency so a USD trip and an INR dinner stay separate ledgers.

Two algorithmic cores:

1. Deterministic remainder-split. When a 100-rupee expense divides three ways, someone has to absorb the extra 0.33. The algorithm assigns the remainder deterministically (by sorted user identifier) so a re-run produces identical splits.
2. Greedy debt simplification (per currency). Max-creditor vs max-debtor matching reduces the total number of transactions in a group. Per-currency means a USD net-creditor doesn't get auto-paid out of an INR pool.

Repayments require creditor approval — the debtor proposes a settlement, the creditor accepts. Both sides see the same notification stream via Supabase realtime.

Key engineering decisions

• Per-currency debt simplification, never mixed-currency. A simplification across currencies would silently re-trade users at exchange rates they did not consent to. Each currency stays isolated.
• Creditor-approved repayments. Prevents the "I paid you back, didn't I?" dispute pattern with an explicit state-machine acceptance step.
• HTTP-only cookies, not localStorage. Sessions stay XSS-resistant.
• Realtime fanout on each expense addition. Participants see updates without a page refresh.

Stack

Next.js 15 with TypeScript, Supabase (Auth, Realtime, Postgres), Tailwind CSS.

Links

• Live demo: bro-split-two.vercel.app
• Repository: github.com/Anshuu2004/BroSplit