This thesis introduces a semi-automated process for optimizing prompt generation using Reinforcement Learning to improve text readability of content generated by Large Language Models. A novel readability metric normalization technique is employed to ensure consistent evaluation across text samples. The research utilizes a distributed system architecture to integrate multiple services, including word pool scraping, prompt generation and text evaluation, enabling scalable and efficient training of an agent. Results indicate significant improvements in text readability, demonstrating the effectiveness of the proposed approach.