AI Powered Personalized Coach for Dance and Fitness Training

Maintaining precise posture and movement align- ment is fundamental in fitness and dance training, where incor- rect technique can result in reduced performance and increased injury risk. Traditional instruction methods and mobile appli- cations provide limited real-time corrective feedback and lack personalization. This paper presents an AI-Powered Personalized Coach, an intelligent system designed to deliver adaptive guidance and instant posture correction using advanced computer vision and deep learning. The framework integrates YOLOv5 for real-time human detection and MediaPipe for 33-point skeletal landmark estimation, enabling high-accuracy motion tracking. Detected joint angles are analyzed through a hybrid rule-based and machine-learning model that classifies posture deviations as correct, incorrect, or unsafe. The system provides immediate feedback via visual overlays and synthesized voice prompts while logging performance metrics for long-term progress evaluation. A reinforcement-style learning loop allows dynamic adaptation of feedback based on user improvement trends. Experimental evaluation demonstrates improved accuracy in posture recognition, reduced feedback latency, and enhanced engagement compared to traditional systems. This work contributes a scalable, low-cost, and domain-flexible framework that bridges the gap between human coaching and intelligent automation in fitness and dance training.