RF-UV-11/Indian-Sign-Language-Recognition-MediaPipe ? reverse-engineered prompt

Build me a Python project for Indian Sign Language recognition using webcam video and saved video files.

I want one part that can go through a folder of sign language videos, detect the hand landmarks with MediaPipe, and save the extracted keypoints along with landmark images so they can be used later for training or analysis. I also want a second part that opens the webcam and recognizes the sign in real time using a trained model from the model folder, then shows the predicted gesture on screen.

Please keep it simple to run, with clear scripts for landmark extraction and live recognition, and make sure the code is organized so I can easily swap in my own model later. If anything needs current setup details or library behavior, look up the latest docs online. Also include any small helper functions needed to keep the code clean and reusable.