As @wasea already stated, we’re using the Document Undestanding Framework as well.
It’s actually really really good