For this you can go with markdown of PDFs. The markdown text will allow you to capture heading for the PDFs content. You guys can use 'Docling' for this purpose as it guarantees all data capturing along with well-structured tables markdown and images references in markdown text.

• Converts any PDF document to JSON or Markdown format.
• Extracts metadata from the document, such as title, authors, references and language.
• Headings will be captured by "##" followed by content within the heading.
• OCR support is also present.
• Integrates easily with LLM app / RAG frameworks.
• Inbuilt Hierarchical Chunking is also supported.

Docling Official Documentation - https://github.com/DS4SD/docling

Alternatively, you can also use PyMuPDF4llm which also does the same mostly but internal order and table markdown could be not as perfect in comparison to Docling.

PyMuPDF4LLM Official Documentation - https://pymupdf.readthedocs.io/en/latest/pymupdf4llm/