datalab-to/surya ? reverse-engineered prompt

I want to build a tool that can look at a document, like a PDF or an image, and intelligently pull out all the information. It needs to do more than just basic text recognition. It should be able to analyze the page's layout, identifying different elements like headers, paragraphs, images, and tables.

A key feature is understanding the correct reading order, so if I give it a newspaper article with multiple columns, it knows how to read it properly. It should also be great at recognizing tables and understanding their row and column structure. This needs to work for a huge range of languages, like Chinese, Arabic, and Hindi, not just English. It would also be amazing if it could recognize and translate complex mathematical equations from a scientific paper into LaTeX.

Please build a simple web app with an interface where I can upload a file and see the analysis visually, with boxes drawn on the document showing the detected text and layout areas.