AI Transparency
How Emvio's AI interviewer works, what it measures, its limitations, and your rights as a candidate.
Employment screening AI is classified as high-risk under EU AI Act 2024/1689. The disclosures below fulfil Art. 13 (transparency), NYC Local Law 144, and Illinois AI Video Interview Act notice requirements.
What this system is and does
Intended purpose
Emvio's AI system — "Riya" — conducts structured voice interviews with job candidates on behalf of hiring companies. It generates a ranked shortlist and a written hiring brief. Riya is intended to supplement, not replace, human judgement in hiring decisions.
What the AI actually measures
Candidates are scored across four axes. Scores range 0–100. Weights are fixed and disclosed below.
Accuracy, depth, and specificity of answers to role-relevant technical questions.
Consistency between answers and behavioural signals, with anti-cheat analysis applied silently.
How closely the candidate's described experience aligns with the job description requirements.
Clarity, structure, and articulation. Not penalised for accent or English proficiency.
Input data
- micVoice audio from the candidate's microphone (streamed live to Deepgram STT)
- descriptionJob description text provided by the hiring company
- attach_fileCV / résumé (if submitted before the interview)
- timerResponse latency timestamps (milliseconds between question end and first word)
Known limitations and risk factors
Audio quality affects STT accuracy
Poor microphone quality or background noise can cause transcription errors. Candidates can request a re-interview if they believe technical problems affected their assessment.
Not validated across all demographic groups
Emvio has not yet conducted statistically significant disparate-impact testing across protected classes. Hiring companies are advised not to rely on AI scores as the sole basis for decisions about protected groups.
Anti-cheat signals are probabilistic, not definitive
High cognitive-latency scores flag possible AI assistance — they do not prove it. A human recruiter reviews all candidates flagged by the anti-cheat engine before any adverse decision is made.
Scoring is role-specific
Scores are calibrated against a specific job description. The same candidate may score differently across different job roles. A low score for one role does not indicate general unsuitability.
Human oversight mechanisms
Request human review
Any candidate whose application was assessed by AI may request a human recruiter to personally review their scores. This is a standing right under GDPR Art. 22 — not a one-off exception. Use the "Request Human Review" button on My Applications.
Final hiring decision is human
Emvio's Terms of Service prohibit companies from making hiring decisions solely on the basis of AI output. The shortlist is a ranked recommendation — the hiring manager is required to make the final call.
Immutable audit trail
Every scoring decision, human review action, and data erasure is recorded in an append-only audit log that cannot be modified (EU AI Act Art. 26). Candidates may request a copy via the Privacy & Data Centre.
Scores are visible to you
After your interview, you can view your scores and the written assessment on your Assessment Results page. The hiring company sees the same breakdown.
What this AI does not do
Does not analyse facial expressions, emotions, or appearance. Camera is used only for presence verification and (Phase 2) gaze anti-cheat — never for sentiment or emotion detection.
Does not use demographic proxies. Race, gender, age, nationality, accent strength, and religion are not inputs to the scoring model.
Does not generate a final hire/no-hire decision. It generates a ranked shortlist and a written rationale — the hiring company decides.
Does not use your interview data to train AI models. Interview transcripts are processed for scoring only and are not used as training data without separate explicit consent.
NYC Local Law 144 — Bias audit summary
New York City Local Law 144 requires employers using Automated Employment Decision Tools (AEDTs) to conduct an annual independent bias audit and make a summary of the results publicly available. Emvio's bias audit covers the CQE scoring system used for candidate shortlisting.
Bias audit methodology
EEOC Four-Fifths (4/5ths) Rule: adverse impact ratio (AIR) = group selection rate ÷ highest group selection rate. AIR below 0.80 indicates potential adverse impact.
Audit status
First independent third-party bias audit scheduled Q4 2026. Automated internal monitoring runs monthly via our disparate impact testing tool.
Protected categories assessed
Race/ethnicity, sex (EEO-1 categories). Emvio does not collect demographic data directly; proxy inference is used for internal monitoring only.
Demographic data
Emvio does not collect race, gender, age, or national origin. These signals are explicitly excluded from all scoring model inputs.
EU AI Act technical documentation
Under EU AI Act Art. 11, Emvio maintains technical documentation for this high-risk AI system covering system architecture, scoring methodology, training data description, performance metrics, risk register, and conformity assessment. This documentation is available to national market surveillance authorities within 10 working days of a request.
Questions about AI decisions on your specific application? Help Centre · Technical documentation requests: [email protected]