Learning engagement assistant (LEA): a multi-agent AI framework for adaptive and personalized learning with simulated student agents

Recent advances in Artificial Intelligence and Large Language Models (LLMs) are enabling adaptive agent systems for personalized learning. However, adoption in higher education remains constrained by variability in course design, learner diversity, and the need for pedagogical alignment and instructor oversight. This paper presents the Learning Engagement Assistant (LEA)-a tri-modal, adaptive AI agent that delivers individualized instruction through integrated Chat, Tutor, and Quiz modes. LEA combines course-specific Retrieval-Augmented Generation (RAG) and Knowledge Component (KC) models to provide contextually grounded instruction and assessment to support scalability across instructional domains. A multi-agent orchestration dynamically adjusts task difficulty and scaffolding using learner performance, cognitive load estimation, zone of proximal development inference, and motivation tracking. The contributions are: (1) a pedagogically grounded orchestration framework integrating knowledge modeling and mastery estimation with tri-modal content generation; (2) a scalable knowledge representation pipeline achieved through modular course RAG knowledge bases and KC models; and (3) an evaluation framework with mode-specific performance metrics and simulated learner agents. Simulation findings demonstrate robust retrieval accuracy, coherent multi-turn tutoring, and adaptive stability across learner profiles and domain content, indicating that LEA can support pedagogical consistency across subject areas and dynamic learner-responsive support.