TutoBot : Un système de tutorat intelligent basé sur les LLMs et une architecture RAG hybride

Abstract

The transition from secondary education to higher education is a critical stage in students’ academic journeys. First-year university students often encounter significant challenges in adapting to new academic regulations, administrative procedures, learning methods, and the overall university environment. Although universities provide access to various information sources, such as institutional websites and administrative documents, this information is often fragmented and difficult to navigate. As a result, students may struggle to obtain timely, accurate, and context-specific guidance during their transition to university life. Recent advances in Large Language Models (LLMs) have created new opportunities for developing Intelligent Tutoring Systems. These models can provide natural-language interactions and personalized assistance; however, their direct use in academic environments remains limited by several challenges, including hallucinations, insufficient institutional grounding, and a lack of access to up-to-date university-specific information. To address these limitations, this work proposes TutoBot, an Intelligent Tutoring System based on a hybrid Retrieval-Augmented Generation (RAG) architecture. The proposed framework integrates a Large Language Model with a hybrid retrieval mechanism connected to a university-specific knowledge base. This knowledge base includes academic regulations, administrative procedures, tutoring-related documents, and curriculum information. By combining retrieval and generation capabilities, TutoBot delivers reliable, context-aware, and traceable responses tailored to students’ academic and institutional needs. An experimental evaluation was conducted to analyze the quality of the generated responses, the relevance of the retrieved sources, and the system’s ability to effectively exploit academic documents. The obtained results confirm the relevance of the proposed approach for improving access to university information through an Intelligent Tutoring System.