IMPLEMENTATION OF A MOBILE CALISTHENICS TRAINING SYSTEM

The Mobile Calisthenics Training App presented in this study addresses the lack of adaptive, personalized coaching in bodyweight fitness by integrating advanced machine learning techniques with a robust software architecture. Grounded in a Design Science Research methodology, the system combines on-device TensorFlow Lite posture assessment, a cloud-hosted Proximal Policy Optimization (PPO) reinforcement learning engine, and real-time event streaming via Kafka to deliver individualized workout plans that evolve based on user performance, fatigue, and preferences. A three-tier architecture ensures resilience through circuit-breaker patterns, load balancing, and secure data handling in MongoDB Atlas, while Firebase services provide authentication, analytics, and push notifications. Comprehensive testing—including unit, integration, security (SAST/DAST), and chaos engineering—validated sub-200 ms inference latency, ≥99.9% uptime, and a 98% code coverage. A four-week pilot with 25 participants demonstrated a 20% increase in workout adherence, a 35% reduction in form errors, and a System Usability Scale score of 84. Ethical considerations, such as differential privacy for posture metrics and GDPR-compliant consent flows, reinforce user trust. This work contributes a production-grade MLOps pipeline for mobile health, an extensible blueprint for adaptive fitness applications, and empirical evidence supporting the efficacy of ML-driven personalization in calisthenics training.