Coaching plays a vital role in improving athletic performance, but access to skilled coaching can be limited by factors such as time, location, or cost. This study explores the development of a coaching feedback system using large language models to provide athletes with consistent and scalable guidance. By employing prompt engineering and iterative design, we tailored our LLM to emulate effective coaching practices, emphasizing external focus, autonomy-supportive language, positive feedback, and actionable advice. A key limitation of the system lies in its inability to directly observe athletes, which is essential for providing personalized, context-aware insights. To address this, future research should explore integrating IMU sensors to enable real-time performance analysis and compare the LLM’s feedback with that of real coaches. Preliminary evaluations with athletes and coaches suggest the system is promising for complementing traditional coaching [1].