Files
Abstract
Large Language Models (LLMs) are a class of AI models trained on vast amounts of text data to understand and generate human language. While they perform well on general tasks, LLMs often face limitations in domain-specific applications due to knowledge gaps and hallucinations. Additionally, their performance declines when tackling complex, multi-step problems. To address these challenges, AI agents, which often incorporate LLMs as core components, are being introduced to handle more sophisticated, goal-oriented tasks. In this study, we leverage both LLMs and AI agents to develop intelligent systems aimed at enhancing undergraduate education in biomechanics courses. To achieve our goal, we construct a dual-module framework to enhance LLM performance in biomechanics educational tasks: 1) We apply Retrieval-Augmented Generation (RAG) to improve the factual accuracy and logical consistency of LLMs’ responses to the conceptual true/false questions. 2)We build a Multi-Agent System (MAS) to solve calculation-oriented problems involving mathematical reasoning and code execution. Specifically, we evaluate the performance of several LLMs, i.e., Llama3.3-70B, Deepseek-R1-Qwen-32B and Qwen-2.5-32B on a biomechanics dataset comprising 100 true/false conceptual questions and tasks requiring equation formulation and solving. Our results show that RAG significantly boosts the performance and stability of LLMs to answer conceptual questions compared to vanilla models. On the other hand, the MAS constructed using multiple LLMs demonstrates its ability to perform multi-step reasoning, execute dynamic code, and generate structured, explainable solutions for tasks requiring calculation. These findings showcase the potential of applying RAG and MAS in improving LLM performance for specialized educational tasks in engineering classes, offering a promising direction for intelligent tutoring for domain-specific knowledge learning in undergraduate education.