A plain account of what Infyva's AI does in an interview, what it does not look at, how people stay in charge, and where our bias testing stands today.
1. How an interview is scored
- Questions. AI writes questions from the job description and, if the hiring company provided one, the candidate's resume. So by default candidates for the same job get related but not identical questions. A hiring company can switch a job to its own fixed question set, and then every candidate gets the same questions.
- Answers. The candidate answers out loud. The recording is transcribed to text.
- Scoring. An AI model scores each transcribed answer against the skills for the job and writes a short explanation of each score.
- Report. The hiring team sees the per-question scores, the evidence, an overall score and any integrity notes, and can listen to the recording.
2. What is and is not evaluated
| Evaluated | Not evaluated |
|---|---|
| What the candidate says: relevance, accuracy, depth and examples, against the job's skills | Appearance, facial expressions or emotions (we do not run emotion recognition) |
| Written or code answers, where the interview has them | Accent, tone of voice or speaking style |
| Integrity notes, shown separately for a person to judge: leaving the tab, pasting, and the optional face presence check if the candidate allowed it | Gender, race, age, disability or any protected characteristic. Optional demographic answers are stored apart and never reach the scoring model or the hiring team. |
3. People make the decisions
Infyva does not reject or advance anyone. A person at the hiring company makes every decision and can see the evidence behind every score. Candidates can ask for a human review: Use "Request a human review" on your interview completion screen. The request goes to the hiring company, which makes the hiring decision. If you closed that screen, email privacy@infyva.ai and we will add it to the same queue.
4. Bias testing: where we are
Infyva has not yet had an independent bias audit, and we do not publish a fairness report today. What exists is adverse impact tooling. When candidates choose to answer the optional demographic questions, our team can calculate, for a company's interviews:
- the selection rate by gender and by race or ethnicity (advanced or hired, out of those decided), and the scoring rate (share scoring above the median);
- the impact ratio for each group against the highest group, flagging any ratio below 0.80 (the EEOC four-fifths rule);
- groups with fewer than 10 people are reported as too small and left out of ratios, rather than guessed.
Results depend on how many candidates answer the optional questions, so small companies may not have enough data. A company that needs these numbers, or data for an independent auditor, can ask legal@infyva.ai.
5. What employers must do
Some laws put duties on the employer, not on Infyva. In summary (this is not legal advice):
- New York City Local Law 144: before using an automated tool to substantially assist a hiring decision for a NYC role, the employer needs an independent bias audit done within the last year, must publish a summary of it, and must notify candidates at least 10 business days before use: that a tool is used, the job qualifications and characteristics it assesses, and how to ask for an alternative process or accommodation if one is available. Infyva's own pre-interview notice does not replace this.
- Illinois AI Video Interview Act: before the interview, tell the candidate AI may be used, explain how it works and what it looks at, and get consent; share recordings only with people who need them; delete a candidate's recordings within 30 days if asked.
- Colorado, the EU AI Act and similar laws treat AI used in hiring as high-risk and add duties such as risk management, notices and human oversight.
- Disability laws (such as the ADA): offer reasonable accommodations, for example more time or another format.
6. What Infyva shows candidates
Before any microphone, camera or question audio is used, candidates see a notice (version 2026-10-17b) and choose:
- Record my spoken answers: Your answers are recorded and transcribed so the hiring team can listen to them next to your results.
- Use AI to score my answers on job skills: AI scores what you say against the skills for this job. It does not score your appearance, voice tone or emotions. A person at the hiring company makes the hiring decision, and you can ask for a human review.
- Optional: face presence check (optional): Your camera checks, inside your browser, whether one person is in view or more than one. It does not identify you, track your eyes or read your expressions, and no video is sent to us or stored. The hiring team only sees when no one, or more than one person, was in view. If you leave this off, your camera is not used.
The notice also explains retention: recordings are deleted after 90 days, and transcripts and reports after 2 years. See the full schedule.