Is it built based on Machine learning model?

Hi thanks for sharing this repo.
I am wondering what's the technical details of this library? For example, does it need calibration? How can the lib map the eye pos to the screen? Thanks!