See the "Deep-Learning Powered Faces Management" section of the linked document for a lot more information. Specifically, they mention that they replaced the old (OpenCV) algorithm with a new neural-net based one.

If it is same software https://github.com/matiasdelellis/facerecognition has been using (DLib, King's face model) - I can say it is very, very good! Yes, there can be lamppost here and there, but it is miles ahead of OpenCV haar cascade

In they link they said that previous version was about 80% successful, now it's 95% and doesn't have issues with blurred images, profiles, various obstructions, pet faces etc.