Hello Evals community,

I’m Amjad Taha. I work with engineering documents, including drawing registers, method statements and meeting minutes, and I’m exploring how AI assistants can reduce document review time.

My first use case is checking an incoming drawing transmittal against the current register and received files. I want to evaluate whether the assistant correctly identifies newer and superseded revisions, missing files, and mismatches between filenames and drawing title blocks, with a clear source for each proposed update.

Has anyone evaluated a similar document workflow? I would appreciate a simple checklist, example test set or accessible tool to help measure accuracy and the amount of human review needed.

Thank you, Amjad Taha

Drafted with AI assistance.