Raghav Kochhar

Jaipur Robotics

• Engineered stateful Apache Beam/Dataflow pipelines to aggregate sensor streams, implementing complex deduplication logic across adjacent gates to resolve multi-camera overlap
• Optimized streaming job costs by right-sizing Dataflow workers, cutting compute spend by 50%
• Built a Cloud Run service for ONNX door-state inference (model trained from scratch, 98% accuracy), handling video processing with FFmpeg and session storage in AlloyDB
• Architected a video-generation platform (FastAPI + Streamlit) that renders timelapse videos from GCS using Cloud Tasks for async processing

Stealth Startup

• Led three production ML systems for traffic analysis, sign language, and emotion classification with 90%+ accuracy
• Deployed containerized solutions on AWS, reducing inference latency by 25%

• Built a real-time sign language recognition system with EfficientNet and LSTM, achieving 95% accuracy
• Deployed the model via a FastAPI backend and Streamlit interface with live webcam inference

• Developed a seven-class emotion recognition LSTM model trained on 1,200 audio samples
• Built a FastAPI inference API and Streamlit UI for real-time and batch audio analysis

• Created a YOLOv8-powered traffic monitoring system achieving 90% mAP with pedestrian and signal detection
• Integrated traffic-light color recognition for smart-city planning insights

• Implemented a No-Code ML workflow platform for data preprocessing, training, and tuning
• Built interactive dashboards for model benchmarking and performance visualization

• Benchmarked Baseline, T-Free, and STELLA on consumer-grade hardware, validating edge feasibility
• Achieved 75% embedding reduction with T-Free while maintaining accuracy

• Presented at the International Conference on Progressive Computational Intelligence; architected a unified healthcare platform
• Designed engagement workflows that reduced clinic no-show rates via automated AI alerts and scheduling integration