The custom agents correctly analyze PDF files from tickets. Although if a PDF file contains form fields which are filled, these form fields are not detected correctly. The AI agent thinks the form fields are all empty although they are filled.
I think there is a transformation missing before you forward the file to the LLM in the background. If you would transform the PDF file into a picture first, that should not be a problem anymore.
Real world examples can be found in Zendesk Support Ticket #14916377

