A Framework for Using LLMs and RAG to Realize the Automatic Generation of Learning Materials from Lecture Slides

Abstract

In real-world education, the use of PowerPoint slides or similar types of files is becoming more popular. Such slides are easy to distribute to students. Some online courses also provide slides to their students. These files typically contain the key points of each lecture. However, such slides may not be sufficient for students. Students often need to find their desired knowledge from reference materials, consume much of their time. With the development of Large Language Models (LLMs), a form of generative AI specialized in text generation, the automatic creation of such educational materials has become possible. In this paper, we present a framework for generating personalized and detailed learning materials using LLMs and Retrieval-Augmented Generation (RAG), addressing the limitations of traditional lecture slides that lack depth. Furthermore, we apply the VARK model to detect students’ learning styles. The learning materials will be generated to accommodate students’ diverse learning styles. The system operates in five steps: (1)extracting a section structure from slides using LLMs like Copilot, (2) teacher review and refinement of the structure, (3) storing reference materials in a vector database using Dify, (4) generating detailed materials with RAG based on the structure and knowledge base, and (5) evaluating the output through teacher checks and learner feedback. The generated content can be further customized into VARK-specific formats, such as diagrams for visual learners or audio for auditory learners. The proposed method demonstrates that combining LLMs with RAG can enhance the quality and adaptability of educational content.