datalab-to/chandra ? reverse-engineered prompt

I want to build an advanced OCR application that can convert images and PDFs into structured text, like Markdown or HTML, while preserving the original layout. It needs to be really good at handling the hard stuff that other tools mess up, like complex tables, forms with handwritten text and checked boxes, and even mathematical equations. It should also be multilingual, supporting a wide range of languages, not just English.

For the user interface, I'd like two ways to use it. First, create a command-line tool that can process a single file or an entire folder of documents. Second, build a simple interactive web app, maybe using Streamlit, where a user can upload a file and see the OCR output right away. It should also be able to extract any images it finds in the documents. You can use a state-of-the-art model from HuggingFace to power this.