Trust Center & Evidence Library
InterviewLens is built upon transparent scientific consensus and strict European AI regulations. We measure observable behaviors—never inferred emotional states.
EU AI Act Article 5(1)(f)
The EU AI Act prohibits AI systems that infer emotions in workplace and education settings. Observing facial movements ("smiling") is legally permitted; diagnosing emotions ("happy, nervous, confident") is banned. InterviewLens enforces this boundary in all prompts and code.
100% On-Device Processing
MediaPipe landmarkers execute directly in your browser using WebAssembly. Coordinate vectors are discarded immediately after computing per-second summary statistics. Raw video frames never leave your local device.
Barrett et al. (2019) Consensus
Leading psychophysiological reviews confirm that facial configurations do not possess one-to-one fingerprints with emotion categories. We coach delivery habits (cadence, eye direction, composure), never your internal character.
Evidence Library (E1–E7)
Every coaching recommendation in your report maps to one of these peer-reviewed scientific findings.
Steady gaze toward the camera/interviewer tends to be rated as engagement and competence.
Direct camera gaze during 60-75% of speaking time correlates with positive observer ratings (r ≈ 0.3-0.4). However, it does not correlate with actual on-the-job performance.
Smiling and head nodding while listening tend to be read as warmth and active listening.
Periodic nodding (2-4 per min) and listening smile signals correlate positively with interpersonal rapport and warmth impressions.
Fluent speech with controlled pacing (130-165 WPM) and minimal fillers is associated with preparedness.
Speech rate fluctuations and filler density (>5 per minute) lead listeners to perceive hesitation. Deliberate 1-2 second pauses at thought boundaries are rated positively.
Vocal variety and pitch modulation tend to be read as energy and communicative interest.
Monotone delivery (pitch standard deviation < 12 Hz) leads to perceived disengagement, while moderate pitch range conveys communicative enthusiasm.
Frequent self-touch episodes (face, hair, neck) tend to be noticed by observers as distraction.
Self-touching (hand to face/neck) increases during moments of high cognitive load or physical arousal, but is frequently an unconscious physical habit rather than an emotional state.
Nonverbal delivery impressions correlate weakly with actual on-the-job performance (r ≈ 0.15).
Delivery shapes how interviewers perceive candidates during the 30-minute interaction, but has almost no predictive validity for whether the candidate will excel at their technical work. Delivery coaching is impression coaching, not competence evaluation.
Facial movements are not reliable or specific indicators of discrete emotional states.
Across cultures and individuals, identical facial configurations (such as smiling or brow furrowing) occur across contradictory internal states (amusement, polite skepticism, concentration, strain). Inferring emotion from computer vision is unscientific and legally prohibited in workplace/education contexts.