This paper presents the development of an LLMpowered educational tool aimed at improving language learning skills such as storytelling, report writing, and essay composition. The tool allows a student to input a question along with his answer, which is then graded by the system, and if the answer is categorized as Bad, the student can request a better answer from our fine-tuned and RL-enhanced system. Traditionally, fine-tuned Large Language Models (LLMs) have been used for domainspecific text generation, where they perform well in structured tasks within specific fields. However, in the context of language learning, these fine-tuned models often struggle to maintain coherence, context relevance, and alignment with educational standards. To address these challenges, we applied Reinforcement Learning (RL) to further optimize the LLM's performance, ensuring that the generated content aligns more closely with human preferences in both quality and relevance. By utilizing locally hosted LLMs like GPT-2 and LLaMA2-7B, our approach offers a cost-effective and scalable solution for making highquality language learning tools more accessible and affordable for students. Our results demonstrate significant performance improvements post-RL tuning, positioning this tool as a practical application for modern educational platforms.