Has Anyone Implemented GenAI-Based Document Understanding in a Real-Time Project?

Incorrect, Generative Classification has been around for quite a while now.
You can see it in the documentation here.