Thanks for implementing that feature @bogdankostic, it saved me from doing it myself :-) Unfortunately I found some wrong page numbers when using the PDFToTextConverter in combination with the PreProcessor and the new add_page_number feature. To be more specific, the pages seem to be incorrect for the files attached. I use the page number of the Document when delivering the search results and found that it's sometimes wrong by +- 3 pages. As an example search for "financial statement" in the tesla-document and you will find a Document on page 81, while the Document is actually on page 84.

tesla_annual_report.pdf
2020_deepmind_anual_report.pdf

I'm not a 100% sure if the bug is easy to fix or if the real solution is to store the page as meta data already when extracting the PDF, instead of doing it during pre processing. Should I open a bug ticket?

HI @brunnurs, thanks for bringing this up! Please open an issue about this and provide some details about the configurations you made for the PDFToTextConverter and the PreProcessor, this would help me immensely to debug this and hopefully provide a fix :)

@bogdankostic I tried to make the bug replicable as easy as possible in the ticket above. let me know if you need more information! Many thanks for your effort