How it works

Apply to shortlist, without a queue.

The whole run takes about twenty minutes of wall-clock time per candidate and none of your team's. Here is every step of it, including the one place the platform acts without asking you.

The run

What happens after someone clicks apply.

Times are elapsed from the application landing. The gap before the call exists so candidates are not phoned the second they hit submit.

  1. 00:00

    Application lands

    The candidate applies from your careers page. Résumé, contact details, and answers to any screening questions you configured arrive together.

  2. 00:05

    Résumé parsed

    Skills, dates, titles and gaps are pulled into structured fields. Parsing is not keyword matching, and the parser reports its own confidence so a file it struggled with is treated as a bad read rather than a weak candidate.

  3. 00:12

    Scored against the role

    Your knockout rules run first as exact matches. Only then does semantic ranking produce a fit score between 0 and 100.

    Runs without you
    Below your fit threshold, the application can be closed here automatically, with no human review. Withheld if the résumé parsed poorly or a recruiter has already acted. Set the threshold to zero to disable it.

  4. 02:00

    The call is placed

    Qualifying candidates are called inside the window you configure. Two retries, then a self-service rescheduling link so the candidate picks a time instead of dropping out.

  5. 15:00

    The interview runs

    Your questions, in your order. The AI identifies itself as software, states the call is recorded, and asks the candidate to confirm before starting. It follows up where an answer comes back thin, and it does not improvise new criteria.

  6. +00:30

    Answers scored

    Every rubric dimension gets a score, and each score carries the part of the answer that produced it. Scoring runs at temperature zero, so the same transcript scores the same way every time.

  7. +00:31

    Routed to a person

    The candidate moves to review with the score, the reasoning, the transcript and the recording attached. The AI recommendation is displayed and never executed.

Your two moments

You step in exactly twice.

Both are judgment calls no system should be making for you. Everything between them is the part that used to eat the week.

  • Setting the rubric

    Before anything runs. The AI measures what you tell it to measure, so the competencies and thresholds you write are the whole ballgame. A rubric containing a proxy for a protected characteristic will be applied faithfully and consistently to the wrong thing.

  • Reading the shortlist

    After the interviews. Candidates arrive ranked with evidence attached. Advancing, holding and rejecting are yours, and every stage change records who made it and when.

Deliberate limits

What it will not do.

These are constraints in the product, not settings. Several of them are the reason the tool can be audited at all.

  • No facial analysis. The interview is voice only, in every state.
  • No scoring of accent, fluency, speech rate, or how confident someone sounded.
  • No personality or emotion inference from the voice.
  • No use of candidate data to train models, ours or a vendor’s.
  • No automated rejection after an interview. Scored candidates go to a person.

Run one role through it.

Post a job, let the interviews run, read the shortlist. Judging it on your own candidates beats judging it on our screenshots.

Get started

Candidates: if an AI interviewed you and you want to know what it recorded or asked, the AI hiring page explains it, and the contact page reaches a person.