An opensource machine learning framework, mediapipe is used to get pose estimation points with the help of openCV. The normalized coordinates of these 33 points across joints of the body is used to train a sequential model with LSTM layers. The training dataset for this  was a custom dataset of 300 videos, constituting to a total of 9000 frames of 4 different actions: pushups, squats, lunges and no-actions (to serve as a null class to avoid false positives). The LSTM model detects exercises, followed by algorithms backed by professional trainers, which give feedback on the form and count repetitions.