Inspiration
Every shared home has the same quiet argument: who paid for what, and who owes whom? Roommates, couples and families split groceries, rent, electricity and takeaway every week. The record lives in a spreadsheet nobody updates or an app nobody opens, because the moment you pay is the moment your hands are full: unpacking shopping in the kitchen, right next to an Echo Show.
We wanted the household's money conversation to happen where the household already talks: out loud. You say who paid and how to split it, and FairShare keeps it exact.
What it does
FairShare is a shared household expense ledger for Alexa+: a self-hosted MCP server, shown on a simulated Echo Show.
- Speak messy, real-life splits. "I spent ₹2,400 on groceries. Split it between me, Priya and Neha, but Neha was away for half the week." FairShare works out that Neha counts as half a share and reads the split back: "Rahul ₹960, Priya ₹960 and Neha ₹480. Shall I save it?"
- Nothing is saved without confirmation. Every new expense is previewed first and saved only when you say "go ahead".
- Knows who's speaking. Alexa+ voice ID is simulated on screen, or people just say "Hi, I'm Priya…". If FairShare doesn't know who's talking, it asks "Who's speaking?" before touching any money.
- Fixes mistakes naturally. "Oops, I told the wrong amount. I paid 550" undoes the ₹500 payment and records ₹550 in one turn. Undo only ever reverts your own last change.
- Every way households split: equal, weighted shares, percentages, exact amounts, and items that belong to only some people.
- Payments, balances, settle-up and monthly totals: "I paid back Rahul 500", "Who owes what?", "What's our total expenditure this month?"
- Voice-first, with a screen. Short spoken replies; cards show the split awaiting confirmation, balances, settle-up and the monthly summary. Like an Echo's Follow-Up Mode, it keeps listening after each reply, so a conversation doesn't need a button press every time.
- Never moves money. It suggests payments; you record them when they happen.
How we built it
The core rule: the AI understands, the server calculates.
Built With
- fastapi
- mcp
- mcp-python-sdk
- nvidia-nemotron
- python
- react
- streamable-http
- typescript
- vite
- web-speech-api
