JobML enables a resume to be simultaneously human-readable and machine-parseable by synchronizing prose with structured data.

Ok screw it, making my own resume a time boxed tool project to add a backrrencing LLM parseable 'JobML' markup language to enable a resume to be both machine and human parseable in the same document.

Ok screw it, making my own resume a time-boxed tool project to add a back-referencing LLM-parseable 'JobML' markup language to enable a resume to be both machine and human parseable in the same document. LIKELY be YAML, but theoretically, it's back referencing, which enables each section to be updated when the prose is (so it auto-syncs). It can use my NER and embedding techniques to keep it in sync (non-LLM required). IDEALLY, you'd have a skills embedding model (or like my lucidResume work a live centroid https://github.com/scottgal/lucidresume based on parsed skills), but failing that, you can put synonyms inline. The idea is that the prose can remain HUMAN-generated, but you get an assist in the same document by seeing what skills you DON'T cover in the prose based on the parsed version. Like Markdown for AIs https://mostlylucid.net/blog/stylobot-release-styloextract, the structured version at the end of the document becomes a live, back-referenced representation of the structured prose with *evidential claims* tied to prose (or projects, GitHub repos, blog articles, etc.). Let's see what a properly structured resume LOOKS like without these dual loss functions playing off against each other. Be honest; here's the human-generated, human-parseable prose, and here's the version for LLMs & screening tools to keep the human prose HUMAN.