Emotion-Aware AI Interview Practice System

Interview performance depends on a candidate’s technical skills, communication clarity, emotional stability, and nonverbal behavior. However, most existing practice platforms use static question sets and ignore real-time emotional or behavioral cues. We present an emotion-aware AI interview practice system that dynamically adapts to the user’s facial expressions, speech clarity, hesitation patterns, and attention cues. Our system integrates MediaPipe face landmark analysis, Whisper/Vosk speech transcription, and open-source large language models (e.g. Mistral [5]) for question generation. Unlike heavy deep models, we use lightweight, rule-based multimodal fusion to enable real-time responsiveness on ordinary hardware. A multi-phase user study shows that participants experienced significant gains in self-awareness, confidence, and communication skills. This work offers a practical, inclusive, and explainable framework for AI-assisted interview training (extended version with in-depth methodology, analysis, and future work).