
2026
Poker Attention
A transformer-based poker AI that rapidly adapts to unknown opponents using attention-based opponent fingerprints. The system features a 4M-parameter model trained via reinforcement learning and supervised learning, with real-time opponent modeling through persistent memory. Built with PyTorch and React, it includes an interactive frontend for visualizing agent behavior, opponent archetypes, and training metrics. The project demonstrates advanced ML engineering with mixed-precision training, efficient inference (~50ms per batch), and comprehensive evaluation tools.



