rubrica

Decision support with a mandatory human gate.

The agent never rejects anyone on its own. It ranks with cited evidence, and stops.

Every candidate here is synthetic. There is no upload, and no real person is scored or contacted.

Start a screening

Why this shape

The EU AI Act lists CV screening as high-risk — Annex III, point 4(a) covers AI systems used "to analyse and filter job applications, and to evaluate candidates." Those obligations take effect on 2 December 2027. This demo is built to that shape now: a written, versioned rubric, a rule code behind every point, and a human gate before anything leaves the system.

How a score is produced

  1. 01

    The model proposes a rubric

    Criteria, weights, thresholds, and what stays explicitly outside the score.

  2. 02

    You edit and approve it

    Until you do, it scores nobody. Editing creates a new version; the old one stays readable.

  3. 03

    The model extracts evidence

    Values with a literal quote from the document. Never a judgement.

  4. 04

    Code produces the score

    A deterministic engine, one rule code per criterion. Same evidence and same rubric, same result — no model call.

  5. 05

    You decide, item by item

    Nothing leaves the system without an explicit approval, and anything done can be undone.