Search

LearnMate: A Personalized Learning Resource Generation System Based on Multi-Agent Collaboration

Author(s)

Xinru Guo†,*, Xiang Li†, Jingyuan Yang†, Qingfeng Zhou

Affiliation(s)

School of Artificial Intelligence and Big Data, Henan University of Technology, Zhengzhou, Henan, China †These authors contributed equally to this work *Corresponding Author

Abstract

Digital teaching platforms today tend to deliver homogenized learning resources with little room for personalized adaptation. Meanwhile, general-purpose large language models frequently produce hallucinated content, which forces teachers to carry heavy repetitive workloads during lesson preparation, while students can hardly find materials that actually match their own proficiency. To address these practical difficulties, we developed LearnMate, a higher-education-oriented platform for curriculum-resource generation and dynamic evaluation that relies on multi-agent collaboration. The system first builds a course-specific RAG knowledge base so that teaching materials remain structurally traceable. It then uses LangGraph to coordinate several agents that generate multimodal resources including courseware, exercises, and teaching videos, and a two-tier quality review mechanism is put in place to mitigate model hallucinations. A six-dimensional learner profile is updated from learning events and used to produce adaptive learning paths and personalized resource recommendations, forming a bidirectional closed loop for teaching and learning. Full-scale practical validation was carried out on the course Big-Data Computing Cluster Technology. In the evaluation, the retrieval hit rate of the course knowledge base reaches 100%, and teachers' lesson-preparation time per lecture drops by 73.3%. The knowledge-absorption rate of underperforming students also improves significantly. The platform supports domestic large-language models and domestic databases, and offers plug-and-play multi-course extension capability. Taken together, the system reduces teachers' lesson-preparation costs and enable hierarchical adaptive learning. It provides a feasible, complete engineering solution for applying large-language models to specialized university courses.

Keywords

Higher Education; Multi-Agent; Personalized Learning; Large-Language Model; RAG; Dynamic Student Profile; Multimodal Resource; Learning Evaluation

References

[1] Wang, Q. Large language models in higher engineering education: Opportunities and risks. Higher Education Quarterly, 2023, Vol.77, Iss.3, pp.341–358. [2] Zhang, L., & Wang, H. Adaptive learning systems in higher education: A systematic review. Computers & Education, 2023, Vol.217, Iss.5, pp.104289–104306. [3] Liu, Y., Chen, H., Zhou, M., et al. Multi-agent framework for personalized tutoring with retrieval-augmented generation. IEEE Access, 2024, Vol.12, pp.58921–58934. [4] Chen, S. Mitigating hallucinations in educational RAG systems. Journal of Educational Technology & Society, 2024, Vol.27, Iss.3, pp.145–162. [5] Zhao, K. Hybrid vector retrieval strategy for educational knowledge bases. Neurocomputing, 2025, Vol.532, pp.118–132. [6] Li, M. Dynamic student profiling for self-adaptive learning paths. Applied Artificial Intelligence, 2025, Vol.39, Iss.2, pp.201–223. [7] Han, W. A survey of adaptive learning platforms based on generative AI. Journal of Computing in Higher Education, 2023, Vol.35, Iss.2, pp.411–437. [8] Garcia, R. Continuous learning assessment supported by AI agents. British Journal of Educational Technology, 2024, Vol.55, Iss.6, pp.2103–2120. [9] Brown, J. Agent-based intelligent tutoring systems: Trends and challenges. International Journal of Artificial Intelligence in Education, 2022, Vol.32, Iss.4, pp.678–705. [10] Yang, T. Human-in-the-loop multi-agent collaboration for curriculum content generation. Expert Systems with Applications, 2025, Vol.247, pp.123215.