Interview Integrity Analyzer

Intro

🚀 About the Project

Inspiration

Remote interviews are now normal, but trust hasn’t caught up. Interviewers struggle to know whether a candidate is genuinely answering or silently relying on AI tools, tab switching, or off-screen help. Most solutions either invade privacy or require heavy backend infrastructure. I wanted to prove that meaningful, ethical interview integrity checks can be done directly in the browser, without recording or storing personal data.

What It Does

Live Interview Integrity Analyzer is a frontend-only system that analyzes interview behavior in real time. It observes multiple client-side signals—eye gaze deviation, face presence, tab switching, copy-paste actions, focus loss duration, and basic audio anomalies—and converts them into a continuously updating Credibility Score.

Instead of accusing candidates, the system provides explainable signals and a visual timeline so interviewers can make informed decisions.

How I Built It

The entire system runs in the browser using modern Web APIs. Webcam and microphone streams are accessed via WebRTC, while MediaPipe FaceMesh and TensorFlow.js are used for real-time facial landmark detection. Browser events like visibilitychange, keyboard shortcuts, and clipboard access are tracked to detect suspicious context switching. A transparent, weighted scoring model combines these signals: Credibility=100−∑(signal_severity×weight)

A timeline panel visualizes all detected events, making the system interpretable rather than a black box.