This thesis explores the development and training of an AI-driven agent within a turn-based fighting game environment, focusing on reinforcement learning techniques. Through rigorous experimentation and analysis, the agent successfully demonstrates proficiency in strategic decision-making and competitive gameplay. Our findings highlight promising performance trends in cumulative rewards, episode lengths, and policy entropy dynamics, underscoring the effectiveness of the reinforcement learning approach. Additionally, we also acknowledge the limitations of the current approach and propose future research directions, including self-play methodologies and integration of advanced techniques like deep reinforcement learning. Overall, this thesis contributes valuable insights into the field of AI-driven gameplay and lays the foundation for further advancements in artificial intelligence.