shirleydhaka/multimodal-rag-provenance ? reverse-engineered prompt

Build me a Python app that checks whether a claim is actually supported by a PDF, including text, tables, and images.

I want to give it a PDF, a user question, and a statement to verify, then have it extract the relevant content, compare the statement against the document, and tell me if the claim is supported, contradicted, or not found. It should also show where the evidence came from in the PDF, like the page number and the exact area on the page, so I can see the grounding clearly.

Please include a simple way to test it with a sample multimodal PDF too, and make sure it can generate a readable report or visual summary of the provenance. If there are any model or package setup details I need, keep them easy to run from the command line, and look up current docs online if you need to.