Building production recommendation systems. Reduced inference latency by 40% through model quantization, caching strategies, and optimized serving infrastructure. Working with TensorRT and ONNX Runtime for model deployment.

Published work on neural architecture search for edge devices. Built reproducible experiment frameworks adopted by 15+ researchers. Focused on efficient model design and automated hyperparameter optimization.

Led backend development for data-intensive web application. Migrated monolithic architecture to microservices, handling 10TB+ daily data volume. Implemented event-driven architecture using Kafka and built real-time analytics pipeline.