This paper proposes an innovative system for optimising yoga posture alignment using computer vision and deep learning. With increasing mental stress and physical ailments in modern society, yoga offers significant benefits for physical and mental well-being. However, mastering proper postures often requires professional guidance, which is not always accessible. Our proposed system utilises the BlazePose model for keypoint detection and integrates a Convolutional Neural Network-Long Short Term Memory deep learning architecture with a rule-based system to provide real-time posture feedback across various age groups and fitness levels. Trained on seven yoga asanas, the system achieved 97.86% accuracy, an F1-score of 0.968, precision of 0.969 and a recall value of 0.967 for Vrksasana, demonstrating its effectiveness in real-time pose correction and accessibility.