cjami/pokerbot-3000

PokerBot 3000 is a live poker game where a human plays against two AI-powered characters at a three-seat Texas Hold'em table. It was built in 24 hours for a hackathon. The AI opponents are named Eliza and Reachy Mini, a small physical robot. The game uses cameras to read playing cards on the table, including the shared community cards and each AI's private hand. A microphone picks up the human player's spoken commands, such as "call", "raise to 100", or "all in". The system transcribes those commands and advances the game accordingly. The two AI opponents make their own decisions, speak out loud during play using synthesized voices, and can display emotion cues. Under the hood, a large AI vision-language model called Gemma 4 31B, running on the Cerebras inference service, handles both card recognition from camera frames and decision-making for the AI players. ElevenLabs provides speech-to-text for the human player's voice and text-to-speech for the agents' spoken responses. The game state, events, and audio are streamed to an operator dashboard in the browser. Private cards are kept hidden from opposing players throughout. The backend is built with Python 3.12 using FastAPI and Uvicorn. The browser interface uses Tailwind CSS and is bundled with esbuild. Running the project requires API keys for both Cerebras and ElevenLabs, which you configure in a local environment file. Setup uses the uv package manager and Node.js for front-end assets, and the whole thing starts with four shell commands. A physical Reachy Mini robot can optionally be connected for gesture and movement responses, though the game runs without it. The repository includes commands for tests, linting, and formatting. It was described as a hackathon project and is not positioned as a production application.