The integration of fitness tracking devices and health monitoring has highlighted the need for fast, accurate, and real-time exercise recognition and feedback. This research proposes a comprehensive system that combines physical exercise recognition using machine learning algorithms with real-time feedback mechanisms via the Mediapipe and Streamlit frameworks. The system addresses two critical aspects of fitness applications: Accurate exercise categorization and real-time user interaction, providing immediate feedback and progress tracking. The study employs Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTM) networks, and Multilayer Perceptrons (MLPs) with pose landmarks from Mediapipe datasets, achieving up to 88% classification accuracy for exercises like push-ups, squats, and jumping jacks. The real-time feedback system, implemented using Streamlit, offers visual and auditory posture corrections, while a Metrics Dashboard tracks repetitions, joint angles, and posture accuracy over time. This integration ensures high precision and user-oriented performance, with applications in fitness, rehabilitation, and sports training.