Poker IA

Why I built this

I am an independent, non-technical creator from Martinique. I learn by building tools that solve problems I can understand and use myself.

Poker IA already existed as a local poker training application before Build Week. It could simulate a table, give strategic advice and preserve decision history. The missing part was simple: after a session, the player still had to search through the history to understand what to practise next.

During OpenAI Build Week, I focused on that gap and built the Session Coach.

What Poker IA does

Poker IA is an offline No-Limit Texas Hold'em training simulator for two to eight players using fictional chips. It records the decisions entered by the player, estimates the available actions with its local strategy engine, and keeps the complete history on the device.

The new Session Coach turns a completed session into:

• a decision-quality score out of 100;
• cumulative estimated EV loss in big blinds;
• the three costliest decisions, linked directly to their detailed replay;
• a short personalised training plan;
• measurable strengths;
• a downloadable Markdown coaching report.

The score is deliberately independent of the financial result. A lucky mistake should not look like a good decision, and a sound decision should not be punished just because the hand was lost.

How I used Codex and GPT-5.6

I used Codex with GPT-5.6 throughout the Build Week extension. It helped me audit the existing FastAPI, React, TypeScript and SQLite architecture; define the new backend contract; implement the coaching aggregation and interface; connect priority decisions to their replays; add the Markdown export and responsible-use messaging; and build the API and browser test coverage.

The most important decisions I kept control of were the outcome-independent score, the local-only privacy model, and the choice to produce a short, understandable training plan instead of another page of raw statistics.

The repository separates the imported pre-hackathon baseline from the Session Coach commit so the new work can be reviewed directly.

Privacy and responsible use

Poker IA does not connect to poker platforms, observe external tables, place bets or use real money. Cards, opponent notes and sessions stay local. The project is a learning simulator, not a promise of profit.

Validation

The current version passes:

• 117 backend tests;
• 38 frontend tests;
• 26 complete Chromium scenarios.

Source and build evidence: https://github.com/Ryan-Charles/poker-ia

Built With

• codex
• fastapi
• gpt-5.6
• playwright
• python
• react
• sqlite
• typescript