Inspiration

It started with a simple goal: make money playing poker. That quickly became a more interesting challenge—could we build a bot that understands probability, reads its opponent, and avoids getting bullied by constant raises?

What it does

PocketPilot plays heads-up short-deck five-card Pot-Limit Omaha. It estimates its chances of winning, tracks an opponent’s betting habits, and chooses whether to fold, call, or raise.

It considers the price of a call and the strength of possible opposing hands, helping it respond to both aggressive bluffers and cautious players.

How we built it

We built PocketPilot in Python using the TAMU Datathon Heads Up starter bundle, with AI-assisted development through Astra.

We combined exact hand evaluation, NumPy-powered simulations, and an opponent model that updates during play. The finished bot runs locally without an API key or internet connection.

Challenges we ran into

Running out of AI credits was one challenge. The bigger engineering challenge was fitting our decisions into an eight-second computation budget per half.

An early test exposed a timeout against a river bluffer. We reduced repeated calculations and made the bot adjust its simulation workload earlier.

Accomplishments that we're proud of

PocketPilot passed all 26 tests. Across 12,000 hands in the final official practice matches, it beat five scripted opponents and drew against itself, with no errors, skipped actions, or timeouts.

We’re especially proud that we found and fixed a problem through testing before submission.

What we learned

Strong poker decisions require more than strong cards. Pot odds, opponent behavior, and uncertainty all matter.

We also learned that testing against yourself checks consistency, while testing against different opponents reveals weaknesses.

What's next for PocketPilot

We want to test against stronger opponents, improve how quickly PocketPilot recognizes changing tactics, and complete the tournament’s container validation.

The goal is to make its decisions more reliable—and eventually find out how it stacks up against other teams.

Built With

Share this project:

Updates

Submission history