Sistema multi-agente basado en LLMs para la recomendación de Recursos Educativos Abiertos en entornos universitarios

Abstract

Abstract: This work presents the design, development, and validation of OPMAS (Open Personalized Multi-Agent System), a multi-agent system based on Large Language Models (LLMs) for personalized recommendation of Open Educational Resources in university environments. Addressing the challenge of academic information overload, we propose a multi-agent architecture that implements centralized orchestration through LangGraph. The system integrates specialized agents (Query, Search, Adaptation, Vector) that collaborate to process queries, retrieve resources from multiple repositories (MERLOT, OER Commons, CORE, YouTube), evaluate pedagogical quality, and adapt content to the student's academic profile. Furthermore, it incorporates European Union competency frameworks (DigComp, EntreComp, GreenComp) as a curricular alignment mechanism and utilizes vector search with embeddings for efficient semantic retrieval. The research adopts an exploratory and descriptive approach, developing a Minimum Viable Product (MVP). Validation through comprehensive testing demonstrated 95% functional success, highlighting its interoperability by interacting with external APIs and vector databases while optimizing costs and technical complexity. This approach demonstrates the feasibility of multi-agent architectures with LLMs for scalable educational personalization.