This project customizes DouZero, a deep RL framework for the Chinese card game Dou Di Zhu, for practical play in the Happy Landlord game client. It uses pre-trained deep learning models (SL, ADP, WP) to analyze game state and recommend optimal plays. The system performs pixel-level screen recognition to detect hand cards, opponent plays, and game state, displaying AI suggestions within the game window for the human player to execute.