Given the high incidence of stroke in Panama, this work presents an intelligent interface for post-stroke rehabilitation, designed for use in clinical limited resource settings. The solution leverages pre-trained 2D pose estimation models, removing the need for body-worn sensors and enabling real-time monitoring. A comparative analysis of five pose estimation models was conducted, evaluating accuracy, robustness to lighting and body position variability, inference time, and resource consumption on mid-range hardware. The best-performing model was integrated into a user-centered interface that includes modules for cognitive assessment, exercise execution, and real-time feedback through visual cues and voice guidance. The results demonstrate the potential of this solution to provide accessible, accurate, and adaptable rehabilitation, promoting treatment adherence and helping to narrow the gap in access to functional recovery services. This work contributes to the development of practical rehabilitation tools for accessible resource settings.