Confidentiality-Preserving Retrieval-Augmented Generation over Educational Resources

Abstract

The evolution of Large Language Models (LLMs) has profoundly transformed access to educational resources, enabling interaction through natural language and fostering new learning modalities. However, the direct use of such models in educational settings raises critical issues related to response verification and the management of sensitive content. In this context, the Retrieval-Augmented Generation (RAG) paradigm represents a promising solution, as it combines the generative capabilities of LLMs with document retrieval from pre-processed knowledge bases, ensuring transparency and source traceability. This work presents a RAG framework designed for education, which integrates retrieval, fusion, and access-control modules to generate reliable responses that comply with users’ authorization levels. Experimental results show that the system enhances the clarity and relevance of the retrieved resources while preserving both equity and confidentiality.