First of all, he should drop Python for anything resource intensive as such a simulation. And then think about how to optimize the algorithm.

If you ever need to edit a PDF that way, just use Inkscape. It is way better than LO draw for that.

It is not a curse. It does exactly what it is intended to do: Create an archive of a document that is universally reproduceable.

It is a very well designed cul-de-sac for exactly this purpose. Using it for anything else is calling for trouble.

The problem lies in the PDFs themselves. In there are objects that represent lines of glyphs. If you are lucky. A conversion tool can guess which of those lines belong together and produce the text.

It cannot know any intentions behind it, though. Take a numbered list. The first line is two line objects: the number plus the . or the ), and the first line of text. The conversion tool can now guess. As the line blocks with the numbers are all left of the line blocks with text, this could be a numbered list. Or it could be a table with two columns. Nothing in the PDF is giving any hints.

And that is the easy part. This assumes that the document either uses default fonts, or keeps its embedded fonts untouched. If they use embedded fonts and a PDF optimizer that only embeds the used characters and renumbers them, any copy or conversion tool is bound to fail.

Same with protected PDFs where you simply cannot copy the text from the start.

And then there are PDFs that just consist of scanned pages. Here you would need an OCR software to get something readable out of them.

PDF is an archival, output format, the end of a process. Not something to work from.

Always preserve the original file. Keep it safe. If you change tools, make sure you have a conversion path into something editable. The PDF is for giving away, nothing else.