Intelligent Tutoring Agent with Retrieval-Augmented Generation: A Case Study of Quality Management System Course

The application of large language models (LLMs) in specialized education presents challenges due to misalignment with domain-specific content and the risk of hallucination. This paper proposes an intelligent tutoring agent powered by Retrieval-Augmented Generation (RAG) to overcome these limitations, focusing on a Quality Management System (QMS) course. By integrating a vector-based semantic retrieval mechanism with LLMs, the system dynamically grounds responses in structured course materials, ensuring accuracy and contextual relevance. Additionally, tailored system-level prompts enhance semantic understanding and improve query rephrasing for more effective retrieval. A case study demonstrates the agent's ability to provide precise conceptual explanations, generate practice questions, and facilitate real-world application of theoretical knowledge, significantly improving student engagement and comprehension. Furthermore, a generalization test highlights existing limitations in adapting to broader contexts, identifying key challenges and future research directions in multimodal retrieval and domain adaptation. Our findings underscore the potential of RAG-based architectures to bridge the gap between general-purpose LLMs and domain-specific education, offering a scalable and adaptable solution for technical learning environments.