LLM resume evaluation — privacy-first by design.
Upload a resume (.txt, .md, or .docx, max 1 MB)
plus evaluation criteria. Before anything reaches the evaluator model, the
pipeline runs a prompt-injection pre-check and a
protected-class scrub (regex baseline + LLM-discovered
terms, applied programmatically) — the model only ever sees scrubbed
text. Each criterion is scored 1–5 by a fresh LLM
context per judgment, with rationale and evidence; inconsistent samples
trigger rubric tightening and a re-score.
Uploads are deleted from disk as soon as the job finishes. Scoring runs with thinking disabled (deterministic, ~5–10 s per sample); a full run typically takes a few minutes.