<p>The present article introduces an innovative method for estimating yoga poses in real-time by combining the accuracy of the MoveNet Thunder model for training with the computational efficiency of the MoveNet Lightning model for inference. Despite the numerous benefits of yoga, performing poses correctly to maximize benefits and minimize injury risk remains a challenge. Our methodology addresses this issue by leveraging deep learning techniques for accurate pose classification of 11 common yoga poses using a dataset of over 3400 images, achieving an accuracy of 99%. The MoveNet Thunder model is employed to train a custom pose classification model on an extensive dataset of yoga images pre-processed using deep learning methods. Following the training phase, the lightweight MoveNet Lightning model is utilized for real-time inference, enabling fast and accurate pose estimation. This hybrid approach overcomes the limitations of longer processing times and inaccurate pose estimation, providing a practical solution for individuals to perform yoga poses more effectively and safely. The potential impact of this research is significant, as it can revolutionize the way yoga is practiced in the digital age, enabling real-time feedback and guidance for correct pose execution. By combining the strengths of accurate pose classification and efficient inference, our approach offers a promising solution for enhancing the yoga experience and promoting overall well-being.</p>