INTEGRATION OF COZE AI MULTI-AGENT VIRTUAL ASSISTANT INTO MOODLE LMS: TECHNICAL ARCHITECTURE ANALYSIS AND TEACHING PERFORMANCE EVALUATION

Tóm tắt

In the context of higher education digital transformation, the application of large language models (LLMs) and Multi-Agent Systems (MAS) is an inevitable trend. This study presents the architecture and empirical evaluation of the successful integration of the Coze AI Virtual Assistant into the Moodle Learning Management System (LMS) at Viet Tri University of Industry (VUI). We focus on designing the Coze Bot with specialized "Skills" (Tools) operating as AI agents to support faculty in three key aspects: (1) Assisting in course content design and development; (2) Automating student frequently asked questions (FAQ) through querying a dedicated Knowledge Base; and (3) Providing summarized analysis of learning activities. The technical architecture employs an API/Chat SDK to embed the domain-specific LLM into the LMS user interface. Empirical results indicate that the solution reduced the time spent by faculty on answering recurring FAQ by 73% (p < 0.01, validated by paired-sample ttest), while enhancing timely 24/7 interaction. This demonstrates the potential of integrating domain-specific knowledge via Retrieval-Augmented Generation (RAG) within a Multi-Agent architecture to significantly improve work efficiency and teaching quality in higher education. Keywords: AI Coze, Virtual Assistant, Moodle LMS, Multi-Agent Architecture, Knowledge Base, Teaching Performance, API Integration.