Stroke remains a leading cause of long-term disability, with significant impacts on motorfunction and quality of life. Traditional rehabilitation methods, while effective, are resource-intensiveand limited in accessibility, particularly in marginalized communities. To address these challenges, thepaper proposes a real-time AI-driven rehabilitation assistant leveraging MediaPipe’s BlazePose framework.This system captures and analyzes user movements, providing features such as pose estimation, repetitioncounter, and real-time feedback to ensure accurate exercise performance. The proposed model was evaluatedon a custom, self-created dataset, designed to mimic real-world scenarios across diverse exercises. Thesystem achieved robust performance, with an aggregate F1-score of 0.8716 and an accuracy of 86.22%,demonstrating its reliability in identifying incorrect forms. The inclusion of text-to-speech feedback and postexerciseanalytics further enhances accessibility and user engagement. By integrating advanced techniqueswith pose estimation, this system offers a scalable, cost-effective solution for stroke rehabilitation, aimingto transform recovery practices and improve accessibility for underserved populations.