This project combines computer vision (ResNet101) with a GPT-style decoder to create an AI agent capable of playing the mobile game King of Glory. The model processes game screen input through a pre-trained ResNet101, then uses a Transformer decoder to generate actions. It requires a physical Android phone with screen mirroring via scrcpy and touch input via minitouch to interact with the game. The author notes the agent currently exhibits problematic behaviors like feeding kills and recommends the successor project which uses reinforcement learning.