AI Explainability Statement

What the AI does

During a session, a speech model converses with the candidate (or supports a human interviewer) using the rubric and question plan you configured. After the session, an assessment model scores each criterion as pass, miss, or insufficient evidence, and must quote the transcript span that supports the call.

What it never does

• It does not analyze faces, clothing, room, or background.
• It does not score accent, tone, pitch, or “confidence” as a vibe.
• It does not output a single 1–10 gut score as the decision.
• It does not auto-advance or auto-reject a candidate. People do.

How you can inspect a score

Open the session report. Each criterion shows the quote, the timestamp, and a confidence note. If the model cannot cite evidence, the criterion is not credited. You can override any recommendation.

Models in use

Live conversation uses OpenAI Realtime. Transcription uses OpenAI Whisper. Assessment uses a documented OpenAI chat model. We will update this page when those providers or model names change.

Humans decide

Hireflo recommends; your team hires. That is policy, not a disclaimer buried at the end of a contract.