I run into the same issue with an internal company RAG, all unstructured data in PDFs but even once converted to markdown, they still need fine-tuning and a lot

I run into the same issue with an internal company RAG, all unstructured data in PDFs but even once converted to markdown, they still need fine-tuning and a lot of manual intervention. It feels like we are inching closer to automating this type of thing, or at the very least brute-forcing it in like the LLM race is trying to do with bigger models and larger contexts. Will have to play with this over a weekend and see what it might help me with :)