Kanishk Varshney

“Multimodal systems for scene-level video search”

Building LLM-driven pipelines for dense video metadata and natural language search.

• Built dense metadata pipelines for scene-level video understanding
• Enabled natural language search across scenes using LLMs
• Structured video content into queryable multimodal representations

“Building and scaling AI/ML systems for satellite data.”

Led ML systems for processing satellite imagery into infrastructure and risk insights.

• Built satellite imagery ML pipelines processing 10K+ images/hour
• Developed vegetation and infrastructure detection models using CV
• Deployed production systems driving $1M+ ARR

“Real-time computer vision for driver safety.”

Built production CV systems for fatigue and distraction detection.

• Built yawn detection system using EfficientNet + LSTM (95%+ accuracy)
• Developed real-time drowsiness detection using eye closure and blink signals
• Engineered seatbelt detection system (98%+ accuracy) robust to lighting conditions