A Modular Multi-Agent Architecture for Hybrid Educational Recommendation Systems Integrating RAG and LLMs

Abstract

Secondary education faces persistent challenges—including heterogeneous learning styles, uneven academic levels, overloaded curricula, and declining student motivation—that hinder scalable personalization. To address these complexities, we designed a modular recommendation system based on a multi-agent architecture, where specialized AI agents collaborate to deliver adaptive learning experiences.

At the core of this system is a personalized learning assistant agent, powered by large language models (LLMs) and a Retrieval-Augmented Generation (RAG) framework. This agent dynamically analyzes student behavior, academic progress, and engagement patterns to recommend tailored content and exercises. It leverages a hybrid recommendation strategy, combining collaborative filtering (based on shared learning trajectories) with content-based filtering (factoring in subject matter, difficulty, and resource type).

Complementing this, a profiling agent continuously updates student profiles—both implicitly through interaction tracking and explicitly via input data—while a clustering agent forms student groups based on learning styles, proficiency levels, and knowledge gaps. These clusters guide the recommendation flow and allow educators to contribute curated resources aligned with each group.

A retrieval agent ensures access to relevant educational materials, while a generation agent produces semantically rich suggestions in real time, thanks to the LLM-RAG integration. The classification engine employs k-Nearest Neighbors (KNN) with Euclidean distance for matching, and Stochastic Gradient Descent (SGD) to optimize prediction accuracy.