This AI first predicts key points or coordinates of different parts of the body(basically where they are present in an image) and then it use another classification model to classify the poses if someone is doing a pose and if AI detects that pose more than 95% probability and then it will notify you are doing correctly(by making virtual skeleton green).