For this hackathon, we built a heads-up poker bot arena where two AI-built bots play each other and improve through repeated head-to-head matches.
How we used Cursor: Cursor was our core AI coding agent. We used it to write both poker bots from scratch, including hand evaluation, pot-odds math, bet sizing, and opponent modeling. We also used it to build the match engine that runs heads-up games, tracks results, and logs every hand. Each time a bot lost, we gave Cursor the hand logs and asked it to find the leaks in that bot’s strategy, such as over-folding to aggression or calling too loosely on the river. Cursor then refactored and tuned the logic. Its multi-file editing let us change strategy modules without breaking the engine.
How we used Grok Bot: Grok Bot automation ran the improvement loop. After each batch of heads-up matches, Grok Bot automatically pulled the results, summarized win rates and key mistakes for each bot, and posted a report. It then kicked off the next round of improvements by handing those findings back to Cursor as targeted prompts. This turned what would have been manual match review into a continuous cycle: play, analyze, improve, replay.
Built With
- fastapi
- python
- uvicorn
