Unstructured-IO/unstructured ? reverse-engineered prompt

I need a tool that can take all sorts of unstructured documents, like PDFs, Word docs, HTML files, and even images with text, and turn them into clean data that's ready to be used by a large language model.

Basically, I want to build a Python library that can ingest a file and automatically parse its contents. It should be smart enough to identify different elements like titles, paragraphs, lists, or tables within the document. The main goal is to convert this messy input into a structured format, like clean text or a JSON object, that I can then easily feed into an AI for things like RAG or fine-tuning. It should act as a pre-processing pipeline for all kinds of documents so I don't have to do the cleaning by hand.