Interview Bar

Preparing for high-stakes technical interviews requires both domain depth and non-verbal communication finesse. Interview Bar bridges both by providing a fully multimodal AI interviewer that evaluates spoken answers, technical accuracy, and facial expressions in real time.

The system integrates a custom LangChain retrieval-augmented generation (RAG) architecture that indexes domain-specific question banks and evaluation criteria. Candidate speech is transcribed using high-precision Speech-to-Text pipelines achieving 95% accuracy, evaluated for semantic relevance, clarity, and completeness in under 2 seconds.

Simultaneously, a lightweight computer vision pipeline built with OpenCV and MediaPipe captures webcam frames to classify facial emotion and engagement levels across 7 classes (neutral, confident, nervous, engaged, confused, stressed, enthusiastic) with 90% precision. The resulting multimodal telemetry is aggregated into an interactive post-session score breakdown.