Submits a public PDF URL and a natural-language question, then returns a text answer based on the document content.