deepseek-ai/DeepSeek-OCR ? reverse-engineered prompt

I want to build an advanced OCR tool with Python. It should be able to take an image or a PDF that a user uploads and extract all the text from it.

This shouldn't just be a plain text dump, though. The tool needs to be smart about understanding the document's structure. It should recognize things like tables, columns, and headings and be able to convert the entire document into clean, formatted markdown. I also want it to handle more complex content, like pulling text from charts and figures inside the document.

Let's use a powerful, open-source model like DeepSeek-OCR to do the actual visual recognition. For the tech stack, we can use FastAPI for the backend API and create a simple web interface with HTML and JavaScript where a user can upload a file and see the extracted markdown output.