docling-project/docling ? reverse-engineered prompt

I need a powerful Python tool to get all my documents ready for an AI application. The main goal is to convert various file types into a clean format that an LLM can understand, like Markdown or JSON.

It should handle all the common stuff—PDFs, DOCX, PPTX, XLSX, HTML—but also more complex things like scanned images with OCR, and even audio files by transcribing them. For PDFs, it needs to be really intelligent about understanding the structure. It shouldn't just dump text; it should correctly preserve the reading order, extract tables accurately, and identify things like code blocks and charts.

I want to be able to run this locally for sensitive data. It should work both as a simple command-line tool for converting single files and as a library I can import into my own Python scripts. Please make sure it has built-in integrations for popular AI frameworks like LangChain and LlamaIndex to make it easy to plug into a RAG pipeline.