How to leverage Azure Computer Vision to create searchable PDFs?

I'm looking into the Azure Cognitive Services / Computer Vision API / Read API to create a searchable PDF. I got as far as retrieving the text data in a PDF.

My specific problem is that I haven't found a good way to enrich the input PDF with the text. My main issue is that I have a bounding box, but I don't have a font size, making it quite hard to really overlay the data on the PDF. In theory, I would have to calculate font size, letter spacing, word spacing and all that to accurately project the text onto the PDF.

I was wondering if anybody here already uses Read API to create searchable PDFs?

I tried using Tesseract as an alternative, and the text is proected fine onto the resulting PDF, but the text quality is really no match to what Read API produces.