Jobbi: application drafts that trace back to approved evidence

Jobbi is an internal tool I built for my own job search. I needed tailored resumes to stay tied to approved experience as the requirements changed from role to role.

I built the evidence catalog, requirement matching, review interface and draft generator. The working product selects approved claims, turns missing evidence into questions and saves a draft with a record of what it used. A reviewer can inspect the support behind the document before deciding to use it.

I separated approved career claims from the vocabulary used to recognize job requirements. Each claim points back to a source and an approved component of the resume. The loader rejects blocked, unverified or uncataloged claims.

The matching layer uses that vocabulary to recognize requirements and rank the relevant approved claims. A requested skill can change which evidence appears first. It cannot create new experience.

If a recognized requirement has no supporting claim, Jobbi creates a question instead of inserting it into the resume. The job detail page shows both the evidence it can use and the information it still needs.

A shared review queue groups the same missing capability across jobs. That keeps one unanswered question from becoming several slightly different drafting problems. Once I add a verified answer to the approved catalog, it can support later work that needs the same fact.

Each generated draft record stores the selected claim IDs, the recognized requirements, unresolved gaps and fingerprints of the source catalog and job input. I can inspect the evidence behind a draft instead of trying to reconstruct how a paragraph appeared.

Generation stops when no approved claims match the posting. Tests cover deterministic selection, unknown claim IDs, unsupported requirements and deduplication of questions across jobs.

Jobbi carries context through research, drafting and review. Preparing material remains separate from an approved submission. That boundary matters when a workflow reaches an external form or needs a human decision.

This is an internal tool I use, not a customer product with a measured conversion result. The concrete deliverable is an evidence map, a review queue and a draft generator that can explain which approved claims it selected. The broader workflow uses agents. This claim-selection step is deterministic.

If your team turns source material into proposals, reports, or other documents, I can build the workflow around what is approved, what is missing, and who reviews the result. Jobbi is my own working example of those decisions. Ask how I would scope your document workflow.