Human pose estimation is crucial for enabling real-time monitoring of physical exercise via the analysis of movement and orientation of the body. However, existing pose estimation techniques are prone to major flaws such as mislocalization of joints, occlusion issues, and mis-recognition of repetition of exercises. Such flaws undermine the efficacy and reliability of fitness tracking systems. In an attempt to address these flaws, the present study proposes a real-time bicep curl tracking system based on OpenCV and MediaPipe. The proposed system is designed to accurately estimate human pose, calculate joint angles, and provide automatic user feedback. One of the system's basic features is that it uses a state-based repetition counter, which improves accuracy in repetition detection by eliminating false positives caused by minor landmark placement variation. The system only detects repetitions when form is proper and range of motion is full. In addition to providing real-time feedback on posture changes and detecting improper exercise form, the system effectively eliminates the risk of injury during the execution of strength training exercises. It provides real-time feedback on posture changes and incorrect exercise form. Through empirical analysis, the system proposed has a remarkable accuracy of 96% in quantifying repetitions, which outperforms the performance of the traditional pose tracking models. The high accuracy verifies the system's robustness as well as its usability in real-world fitness applications. Findings indicate that the integration of AI-driven pose estimation and feedback mechanisms can potentially make personalized fitness training much more effective. Together with real-time correction and individualized data, these technologies can improve efficiency in training while motivating safer training habits. This work contributes to the growing field of AI-driven health and fitness technology and opens the door to more advanced and responsive physical activity monitoring devices