Chipzy

01The problem

A home game is easy right up until it ends. Everyone counts their chips, someone adds up the buy-ins, and the two numbers don't match. A stack gets miscounted, a rebuy goes unrecorded, chips walk off with someone who left early. What follows is a group chat, a calculator, and a final number nobody fully trusts.

Paying up made it messier. Some players hand over cash on the spot, others transfer later, so "who owes whom" is never one clean list. Debts carry between nights. And on the odd trip abroad the game runs in another currency, while everyone still thinks in shekels.

Chipzy started as a C# console app I wrote for my own games, grew into my software engineering final project, and kept growing because we kept using it. The goal became one app that could run the game and close the books: deal the hands, track every session, and settle the night exactly.

02The approach

The core is a real-time table engine. A FastAPI backend holds authoritative game state; clients sync over WebSockets so every seat sees chip movement, blinds, and actions instantly. An admin can correct the table mid-hand, because real home games have real fat-finger moments.

When a game ends, the result imports straight into a cash session, so playing and bookkeeping are one flow instead of two apps. Each session then feeds a permanent record: profit and loss per player across months, who you have played the most, and which debts are still open from previous nights.

• Chip discrepancy solver. When the count is off, the app proposes a resolution (spread proportionally across each winner's win, among other rules) and shows exactly what it adjusted, instead of leaving the table to argue about it.
• Cash-aware settlement. Mark who is paying in physical bills, down to the denominations, and the remainder resolves into the fewest possible transfers.
• Debts in one currency. Play a session abroad in euros or dollars and every debt still lands in your own default currency, so the running balance between friends stays in one unit.
• Equity tools for quick in-the-moment analysis across NLH, PLO4, and PLO5.
• Game organizing over WhatsApp via n8n: send invites to the group where games actually get arranged, collect who is in, and post the results back when the night is done.
• An AI layer on Gemini that writes persona-driven recaps of a night and coaches saved hands street by street with RAG, grading each decision and naming the biggest leak.

03The most configurable table I could build

Every home game has house rules, and the apps I tried all forced their own. So the table creation screen exposes everything: twelve poker variants, dealer's choice, bomb pots and Chocoloko as votable special games, straddles, run-it-multiple-times, time banks, join fees, and per-table timer behaviour. Set it up the way your table actually plays.

04Architecture

Authoritative state on the server, thin real-time clients, and an AI layer that reads what the app already recorded:

05Closing the books

This is the part the app was born for. A real eight-player night: the chip count came up short, so the solver spread the shortfall proportionally across the winners and showed each adjustment. Two players paid in cash, which the settlement takes into account, and the rest resolved into a minimal set of transfers. The recap at the bottom is generated by the AI layer.

06Where it stands

Chipzy is in real use. My friends and I run live and online home games on it as a PWA. It is the project that taught me the most about real-time state, keeping clients honest against a single source of truth, and the unglamorous edge cases that only surface once actual money is on the line.