Large Language Model-Based Sport Coaching System Using Retrieval-Augmented Generation and User Models

Large language models (LLM) have advanced at a high pace in recent years. By using big datasets, they are capable of understanding and generating human-like language. Models like OpenAI’s generative pre-trained transformer (GPT) use deep learning techniques to produce relevant responses in different fields. These models generate natural language applicable for use by artificial agents, such as social robots or chatbots. The challenge now is to personalize these responses for individual users. For this, user models must be used to capture user preferences and behaviors and offer a solution to this challenge. This study designed two LLM swimming coaching systems, both incorporating user models, with one system additionally utilizing a Retrieval-Augmented Generation (RAG) system. RAG increase the quality of the output of a LLM by leveraging contextual or real-world knowledge. Over a three-week period, these systems provided guidance and feedback to improve the swimming performance. Our results showed that participants using the LLM system with RAG significantly enhance their freestyle stroke technique, as evidenced by a reduction in the number of strokes needed to swim a 25 meters lap. This demonstrates the potential of integrating LLMs with RAG and user models to improve personalized coaching in sports.