Not a general tool but arxiv-vanity - which produces webpages of articles submitted to arxiv - works by parsing the source code that's submitted along with the

Not a general tool but arxiv-vanity - which produces webpages of articles submitted to arxiv - works by parsing the source code that's submitted along with the PDF. You can probably use this data to train a model that converts between pdf, tex, and html.