Autonomous Poker Bot

The engine implements legal action validation, street progression, blind posting, and pot tracking. The evaluator supports hand ranking and equity proxies with correctness-focused tests. The ML layer offers both a lightweight RandomForest baseline and a PyTorch MLP with a sizing head.

Key Features

• Rule-consistent game flow (preflop → showdown) with action validation
• Fast 5–7 card evaluation with duplicate-card detection and kicker tie-breaks
• Neural and classic ML predictors with legal-action masking
• Hand history parsing and feature extraction for training datasets
• Extensive tests: unit, integration, and scenario-based

How I Built It

Core in Python. PyTorch for neural models; scikit-learn for the baseline. SQLAlchemy schemas for data storage. Tests via pytest. The code emphasizes clarity, correctness, and forward-compatibility in data schemas and APIs.

What I Learned

• When to play perfect math vs. exploiting weak players
• Processing millions of poker hands without crashing servers
• Making split-second decisions under pressure
• Managing poker variance to stay profitable long-term

Results

• Test suite: 100 passed
• Recent engine fixes: raise semantics, betting-round completion, total pot tracking
• Evaluator upgrades: suit handling (int/char), preflop strength heuristic, compare helper
• ML robustness: tensor/ndarray inputs, zero-amount safety fallback, class-prob alignment
• Data pipeline: flexible GameContext, resilient hero/action extraction