Dialogical Learning Support in RAG-Based E-Learning

The increasing use of large language models in e-learning environments has created new opportunities for natural language interaction between learners and intelligent systems. However, it has also raised important questions about reliability, transparency, and pedagogical appropriateness. Generative AI systems often depend on heterogeneous and insufficiently validated data sources, which can be problematic in educational contexts where accuracy and trust are essential. This paper presents a web-based learning platform built on a Retrieval-Augmented Generation (RAG) architecture, designed to support dialogical learning. Rather than treating learner questions as isolated inputs, the platform views learning as an ongoing dialogue, preserving context across interactions and grounding responses in curated and validated educational materials. The system uses a modular web-based design that clearly separates content management, retrieval and generation, and dialogue handling. This modularity enables the integration of various generative models – both open-source and commercial – and supports deployment in real e-learning environments without requiring local installation. A representative use case demonstrates how the platform can support learning at the university level. Overall, the study shows how dialogically grounded RAG-based systems can improve transparency, contextual coherence, and pedagogical value in AI-supported e-learning.