Design of IoT Architecture and LLM Model for Personalized Training Recommendations for Athletes

Razo-Ballon H.; Ticona-Esquivel R.; Montalvo P. · Communications in Computer and Information Science (2025) · DOI: 10.1007/978-3-031-83210-9_23

Resumen

We propose an architecture of Internet of Things (IoT) and a Large Language Model (LLM) to enhance the training and performance of athletes globally by creating personalized training routines based on their distance goals and physiological and environmental data. Originally motivated by the declining athletic performance at a national level, this work provides a comprehensive system that collects real-time physiological and environmental data through smartwatches, enabling personalized training guidance suitable for athletes worldwide, regardless of their discipline or proficiency level. Our main contribution is the design of an IoT architecture paired with an LLM model, specifically the Llama-2-7b model, which we fine-tuned with diverse training routines to offer tailored recommendations based on an athlete’s specific distance goal. The seamless integration of IoT devices and LLM aims to enhance the overall training experience by leveraging real-time data for precise and adaptive training plans. Experimental results demonstrate the LLM’s ability to generate daily, adaptive training plans that show promise in comparison to the non-tuned model. This integrative approach aims not only to revitalize athletics but also to provide a data-driven foundation for elevated athletic performance on a global scale. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

Instituciones: Universidad Peruana de Ciencias Aplicadas