When you upload a resume, Chosen runs it through an LLM that pulls out structured fields and writes them onto the candidate record. You get a filled-in profile instead of a file you have to read. It's right most of the time, and you can fix anything it isn't.

What gets extracted

The parser reads the resume text and fills in:

Name, email, phone — the contact basics.
Current title and company — where they are now.
Work history — past roles with companies and dates.
Education — degrees and schools.
Skills — what they list.

Those land in the standard candidate fields. Any custom fields you've set to auto-fill get extracted in the same pass — Chosen reads each field's description to know what to look for.

How parsing runs

Parsing happens in the background, not while you wait. A resume moves through a few states:

Queued — uploaded, waiting its turn.
Running — the LLM is reading it.
Succeeded — fields are written to the candidate.
Failed — something went wrong; the candidate exists but its fields are empty.

For a single upload this is quick. For a bulk batch, candidates appear right away and their fields fill in as each resume finishes — the table shows a placeholder on cells still waiting.

Fixing what it got wrong

The parser is good, not perfect. Creative job titles ("Chief Vibe Officer") and dense academic CVs are where it's most likely to miss.

Open the candidate's detail panel.
Edit any field directly — your correction is what everyone sees from then on.
To re-extract everything from the original file, use Reparse.

Reparse re-runs the parser against the source resume. Reach for it after you've added a new auto-fill custom field, or when a search isn't finding a candidate you'd expect — a hand-edit fixes the visible field but a reparse refreshes everything Chosen searches on.

When it gets it wrong

Two non-obvious behaviors worth knowing:

The same file twice is silently deduped. Re-upload a resume Chosen has already seen and it matches the file by content and reuses the existing candidate — no prompt, no duplicate. A different resume for the same person makes a second record. See duplicates and merging.
Very long resumes lose their tail. Parsing reads roughly the first 35,000 characters. A normal one- or two-page resume is nowhere near that. A fifteen-page CV is, and anything past the cutoff — often the oldest roles — won't be extracted.

Parsed data also feeds the match rating, so a clean parse makes that score more reliable too.