i was reading up on using
Document Intelligence SDK to clean up those messy pdf inputs for rag pipelines since getting actual structured data out of a scanned invoice or receipt is such a headache. it's pretty cool how they have these prebuilt models like prebuilt-layout that can just spit out markdown, but i'm skeptical about the effort needed for custom extraction when you're dealing with weirdly formatted templates.
>Most "RAG over PDFs" pipelines have a step nobody talks about much: something has to turn a scanned invoice, a multi-column contract, or a photographed receipt into text a model can actually reason over.it seems like a lot of the heavy lifting is just managing those long-running operations with the poller, which sounds like a fun way to introduce latency if you aren't careful. i wonder how much the classifiers actually help when you're hitting it with a huge mix of unknown document types at once?
article:
https://dzone.com/articles/microsoft-foundry-document-intelligence-sdk