allenai/olmocr ? reverse-engineered prompt

Hey, can you build me a tool for processing documents? I need something that can take a PDF—or even an image file like a PNG or JPEG—and convert it into clean, well-formatted Markdown.

It should be pretty smart and able to handle complex layouts with multiple columns, tables, and even stuff like mathematical equations. It's important that it automatically strips out repeating headers and footers to keep the main content clean. The goal is to get the text in a natural reading order, just like a person would read the page, not all jumbled up.

I know this will probably require a powerful AI model and a GPU to run. Please set the whole thing up in a Docker container to handle all the tricky dependencies. The final tool should be a simple script I can use to process a folder of documents and get the converted text files.