Consider using the Structure Tree to view / work.
That is what AT relies on.
From your screen capture you show incorrect tagging.
(i.e., "Lbl" usage is incorrect).
Consider using ISO 14289 (PDF/UA ISO Standard) & ISO 32000 (PDF ISO Standard) Clause 14 as your "build it" references.
ISO 14289 is available from AIIM for $15.
ISO 32000 Part I can be purchased or download the document from Adobe.
We are not building a document. Instead, we have a pdf document which is converted from LaTex and we are making it accessible for vision disability. So there are instances where we have x cube or mathematical expression or matrix within the paragraph which the screen reader is not reading as it should.
The tags that the screen captures are generated by pdf acrobat itself.
Our concern is how to make such instances accessible.
I also want to inform you of our approach. We convert the mathematical expression as a figure and put alternate text in it.
The "auto-tag" "wizard" of Acrobat Pro is ok (sort of) for simple content.
Regardless, what the screen capture shows in incorrect.
"Our concern is how to make..."
Go back to the basics (start with the two ISO Standards).
Success is predicated on having an knowledgeable/experienced "warm-body" working the PDF.
Accessible PDF *is* PDF/UA compliant PDF; anything else isn't.
While software tools provide much assistance there is no software that "does the job".
Other reference information you want is available at AIIM's website, the PDF Association's website, and the PDF Accessibility Checker site.
Netcentric out of Canada has an enterprise product that can do much but again the knowledgeable/experience human operator is still required.