How Cavenue scores your interview

You start from a job posting: paste a link, write or paste the details, or pick one from the job search. Then you choose the rounds (HR, technical, manager), the language, and whether to speak or type.

The interviewer receives the posting and, if you’ve filled them in, your CV and profile, so its questions fit both the role and your background.

In a voice interview you talk with the interviewer in real time. It runs on OpenAI’s realtime voice model, and your speech is turned into text as you go. Your audio isn’t stored; only the written transcript is saved.

In a written interview you answer in a chat. The questions come from an Anthropic Claude model, with the same posting and profile behind them.

When the interview ends, the whole transcript goes to an Anthropic Claude model together with a fixed scoring guide. Only what you said counts as evidence; the interviewer’s words never do. The guide’s rules:

• Every score rests on evidence from your answers. Missing evidence means a low score, not a neutral one.
• Concrete examples, your own part in them (not just “we”) and measurable results raise a score.
• An area you never really answered scores 0 to 10. Greetings, “yes” and filler don’t count, and a vague one-sentence answer scores 30 at most.
• Contradictions and off-topic answers lower the score.
• Scores are meant to be calibrated and precise, not rounded: most candidates land between 40 and 80.

Your overall score from 0 to 100 is the weighted average of six criteria, each scored from 0 to 100:

• Role knowledge25%
• Problem solving20%
• Communication15%
• Leadership & ownership15%
• Collaboration15%
• Proactivity10%

For each criterion, the report quotes or sums up the evidence it found in your answers and gives one concrete tip.

Each round you choose is also scored on four criteria of its own, using only its part of the transcript. A round’s score is the average of its four.

• HR interview
  • Motivation
  • Experience fit
  • Communication
  • Culture fit
• Technical interview
  • Technical depth
  • Problem solving
  • Trade-off analysis
  • Hands-on experience
• Manager interview
  • Leadership
  • Decision-making
  • Stakeholder management
  • Team building

Four views of the same interview, each from 0 to 100:

• Job readiness
• Interview performance
• Growth potential
• Culture contribution

A recommendation on a six-step scale, from “Strong no hire” to “Strong hire”.

1. Strong no hire
2. No hire
3. Lean no hire
4. Lean hire
5. Hire
6. Strong hire

Strengths, areas to improve, any red flags and an action plan of three to five next steps.

Everything is written in the language of your interview.

• It isn’t a hiring decision or a prediction of one. Real interviewers weigh things Cavenue never sees.
• It reads words, not your voice: tone, accent, pace and body language aren’t scored.
• AI can misjudge an answer, and a transcription mistake can change what the evaluator reads.
• If a transcript is too long to score in one go, turns from its middle are left out, and marked as left out, so the evaluation fits.
• Scores from different postings aren’t directly comparable: each one measures your answers against that role.

If a score or a report looks wrong, tell us from the Support page, with the interview’s date and role. We read every message and correct what we can confirm.