AI Presentation Evaluator

A corporate training client needed an automated, scalable way to evaluate and coach employees on their public speaking skills. Our AI engineering team at NirmataAI built a comprehensive video analysis pipeline that processes raw video uploads to extract multi-modal performance metrics.

We orchestrated a complex serverless pipeline utilizing FFmpeg for audio/frame extraction, OpenAI's Whisper API for precise speech-to-text and filler word detection, and MediaPipe for advanced computer vision tasks like facial landmark detection and eye contact tracking. By aggregating these distinct data streams (confidence, clarity, engagement, and gaze direction), our proprietary scoring engine calculates a weighted composite score.

Finally, we integrated Anthropic's Claude Sonnet to synthesize the raw metric data into actionable, highly personalized coaching feedback. The resulting platform delivers enterprise-grade, unbiased presentation coaching automatically, revolutionizing how the client handles professional development.