PORKER

What It Does

PORKER is an app that predicts the best move in four-card Omaha Hold’em.
When you enter the current game state, it estimates the strongest action for that situation
and presents it as a practical decision aid.
The app also includes CPU matches powered by the same prediction AI,
so it works both as an analysis tool and as a place to practice actual play.

What We Worked On

• Handling the much larger decision space created by four-card Omaha hands
• Combining best-move prediction with playable CPU matches in one application
• Experimenting with both reinforcement learning and supervised learning for the AI
• Reducing manual input effort through computer vision with OpenCV and OCR

Why We Made It

We chose Omaha because it is harder than Texas Hold’em due to the larger hand size,
and because strong AI for Texas already existed.
The project was also a technical challenge: we built the Omaha engine ourselves
in order to support both training and CPU play.

We did try reinforcement learning, but at the time it did not produce an AI that was meaningfully strong.
What worked better was supervised learning using professional match records that we were able to obtain.
That approach led to an AI with much more practical strength,
and gradually shifted the project from an experiment in learning methods
into something closer to a usable decision-support app.

Entering poker situations by hand is tedious, so we also explored whether image recognition could help.
This was before today’s more capable vision models were available,
so we relied on OpenCV and OCR to extract information from images.
The accuracy was limited, but it was an important attempt to improve usability beyond the core AI itself.