Why the interview opens with a line from the candidate's own CV

Candidates arrive at an AI interview expecting a form that talks. The fastest way to get a useful conversation is to disprove that in the opening line.

A generic opener gets a generic answer

"Tell me about yourself" signals that nothing you wrote was read. People respond in kind, with the rehearsed paragraph they use everywhere, and the next fifteen minutes are spent trying to get underneath it.

The avatar instead opens by referencing a concrete detail from that candidate's CV. It is a small thing that changes the register immediately: this is a conversation about my actual work, so I should talk about my actual work.

Follow-ups react to the answer given

There is no fixed question list. The interview runs ten to fifteen minutes, one question at a time, and each follow-up responds to what was actually said rather than advancing through a script. A candidate who mentions a migration gets asked about the migration.

That is also what makes the scorecard worth reading. Every criterion carries the passage from the transcript that produced it, so a hiring manager can disagree with a number and point at the reason.

What a candidate cannot do

An interview nobody watches live is only useful if the record cannot be edited. The transcript comes from the session itself, not from the candidate's browser. The invitation link is genuinely single-use and expires after seven days, so a second attempt is refused rather than quietly starting a fresh session. And CVs, job descriptions and transcripts are fenced and marked to the model as data, never as instructions.

The screening that comes first

Before any of this, each CV is scored 0 to 100 against four to six weighted criteria that Persohap drafts from the job description and a recruiter then edits. The screening prompt instructs the model to disregard name, gender, age and origin.

None of that is a decision. The score ranks the list; a person decides who gets invited.