RAGyverse: A Framework for a RAG-based Interactive Virtual Tutor in Immersive VR

<p>Traditional e-learning platforms often fail to provide the personalized, interactive, and engaging experiences necessary for effective learning. This paper introduces RAGyverse, a novel framework for a multimodal AI-powered tutoring system designed to address these shortcomings. The system’s core is a Knowledge Graph-enhanced Retrieval-Augmented Generation (KG-RAG) pipeline, which grounds responses in factual course material, integrated within an immersive Virtual Reality (VR) environment. A pre-study with potential users confirmed the demand for such interactive features and informed the frame work’s design. To validate our approach, we conducted a focused technical evaluation of the RAG pipeline’s core components, employing an LLM-as-a-judge protocol. The results reveal a critical dependency: the system’s performance is contingent on the quality of its retrieval stage. Successful retrieval leads to high-quality, faithful answers, but failures in retrieval cause the generation component to produce unfaithful, hallucinated responses. This pinpoints retrieval, not generation, as the primary bottleneck. This paper details the framework’s architecture, presents a critical analysis of the RAG component’s performance, and provides a clear roadmap for developing more robust and reliable AI tutoring systems.</p>