What the AI does in an interview, and what it never does
In Vettasy the AI drafts, and a person decides. What the AI writes, where it runs, and the list of things it never does, from ranking to reading emotions.
The AI drafts the scorecard with quotes, summarises the CV, suggests follow-ups and drafts the feedback. It runs on Claude through Amazon Bedrock in Frankfurt. It never rejects, ranks or shortlists a candidate, never saves a score without a person confirming it, never recognises emotions, and never trains on customer data. A person confirms every score.
AI in a Vettasy interview has one job: to draft, so that a person can decide faster and with the evidence in front of them. It never makes the decision. This is what it writes, where it runs, and the explicit list of what it never does.
What the AI drafts
The AI produces drafts, each labelled as a draft until a person confirms it:
- Scorecards. It proposes a score for each competency, with a quote and a timecode from the transcript, so every number points to a moment in the recording.
- A CV summary. It summarises the candidate's CV and the plan for the interviewer before the call.
- Follow-up prompts. During the call it suggests follow-up questions, as suggestions, not a script.
- Feedback. It drafts the written feedback from the confirmed scorecard, which the interviewer edits and sends.
- Highlights. It selects moments for a short summary for the hiring manager.
Where it runs
The AI analysis runs on Anthropic's Claude through Amazon Bedrock in Frankfurt. Customer data is never used to train models. The interview material is processed to produce the draft in front of the interviewer, not to improve a model somewhere else.
What it never does
The list of what the AI does not do is as important as what it does, and it is fixed in every version:
- It never rejects, ranks or shortlists candidates.
- It never saves a score without a person confirming it.
- It never recognises emotions, and never identifies a person by their face. The liveness check keeps no face template.
- It never analyses tone of voice, gaze or posture.
- It never trains on customer data.
A person decides
Every score is confirmed by a person, under the interviewer's name, and the decision log records who decided, when and on which evidence. A signal or a draft under review cannot be grounds for a decision until the person closes it. This is the design the EU AI Act asks for in high-risk uses, human oversight, disclosure and logging, and it is built in now rather than added in 2027. The candidate is told AI is used, in the invitation, on the consent screen and in the app, and every AI output is marked as a draft until a person signs it.
Sources
- Vettasy, AI Notice and Trust center. See also the EU AI Act and hiring.
Written by the Vettasy teamVettasy is a desktop application for job interviews. Employers run structured, recorded interviews in it with verified participants. It is free for candidates. Windows and macOS.
