datalab-to/marker ? reverse-engineered prompt

I want to build a powerful document conversion tool in Python. The main idea is to take various file types—like PDFs, Word docs, PowerPoints, and even images—and convert them into clean, usable markdown and JSON. It should be fast and really accurate.

The quality of the conversion is key. The tool needs to correctly handle tables, forms, math equations, and code blocks. It should also be smart enough to identify and remove artifacts like page headers and footers. Please also make sure it extracts and saves any images it finds in the documents.

I'd like a couple of ways to use it. First, create a simple interactive web app, maybe with Streamlit, where I can upload a file and see the converted output. Second, build a command-line tool that can process an entire folder of documents at once. For the highest accuracy, include a setting to use an LLM like Gemini to help clean up the output, especially for tricky tables.