PDF parser

[swaroopgv]

Question

My Requirement is to parse technical documentations, which may contain tables , meaningful images and provide the contents in JSON format(Since it the suggested format for any LLM models ??)

Currently i have

1. converted PDF to markdown , Since markdown currently does'nt preserve hierarchy, I have written an helper function which takes markdown file , here based on number in the section name , it sorts things and put everything into JSON format and dump to file
   (I hope this is the right approach, if there any better ways please suggest)

Now
2. Here also need to read the tables and convert them to JSON.
3. Take pictures and run it on computer vision model to get extract information about the pictures.

Questions

1. what are the available options for this?
2. if there are any other better approach please suggest.

[dolfim-ibm]

The Docling output is was we call DoclingDocument. You can save it as JSON without losing any content, and later use for all your use cases.

I wouldn't anyway say that the JSON output is the suggested for LLM models. We usually suggest: 1) when converting, store and keep the JSON output, 2) in your downstream pipeline, load the JSON document (parse it with the docling-core library) and export the document to the format needed for the current task, or chunk it directly.

Learn more on serializations and how to customize it here: https://docling-project.github.io/docling/concepts/serialization/