What it is
maxwell-em-processor is the pipeline that turns scanned volumes of Maxwell’s electromagnetic treatises into structured, navigable content. OCR, optional AI classification, hierarchical mapping — then you have articles, not a wall of glyphs.
Why it matters
This is where most “digitize the Treatise” projects stall. You get text. You still do not know where article 528 starts. I split this repo from the Python library on purpose: an OCR retry should not live next to Faraday induction.
How it works
The path the README describes: input PDFs → Mathpix OCR → table-of-contents analysis → optional classification → organized output, keeping Maxwell’s own hierarchy of volume, part, chapter, and article.
In the longer writeup I also noted practical details that belong to this stage: caching so a long rerun does not bill you twice, retry plus a circuit breaker, and a --simple mode if you do not want the classifier. The 1873 scans are on Archive.org; Maxwell’s wording is public domain.
Status
Public at antmikinka/maxwell-em-processor. It feeds the edition layer and, downstream, maxwell-modernized.