Large Language Model–Driven Intelligent Educational Agents for Adaptive Learning: A Systematic Review

This paper presents a comprehensive review of Intelligent Educational Agents (IEAs) developed using Large Language Models (LLMs), the LangChain framework, and Retrieval-Augmented Generation (RAG) techniques for adaptive and personalized learning. The primary objective of this review is to analyze recent advancements in AI-driven educational systems and evaluate how modern architectures overcome the limitations of traditional rule-based and static learning platforms. The study systematically examines existing literature on LLM-based tutoring systems, multi-embedding semantic retrieval, and adaptive difficulty tracking mechanisms used to personalize learning experiences. The reviewed studies indicate that integrating RAG with multi-source embedding’s significantly improves contextual accuracy, response relevance, and knowledge grounding compared to single-embedding or standalone LLM approaches. Furthermore, adaptive learning strategies based on student performance metrics such as accuracy, response time, and hint usage enhance learner engagement and knowledge retention. This review underscores the significance of combining LLMs, LangChain, and adaptive RAG frameworks to build next-generation intelligent educational systems capable of delivering personalized, context-aware, and effective learning support.