Suvendu is a Sr Machine Learning Engineer at Amazon, where he specializes in advancing supervised fine-tuning and RLHF pipelines for large language models ranging from 7B to 470B parameters. With more than 15 years of experience, Suvendu has architected scalable distributed-training systems using Megatron-LM 3D parallelism, DeepSpeed ZeRO, PyTorch FSDP, and AWS Trainium—consistently reducing training costs and latency while maintaining model quality.
His MLOps expertise spans SageMaker, MLflow, and TensorRT-based on-device inference, where he has delivered 3× throughput gains and 30% latency reductions in production workloads. Suvendu has also built real-time recommendation systems at HBO Max and predictive maintenance platforms at Equinix. An active open-source contributor and author of a widely adopted MLOps framework on AWS’s GitHub, he regularly mentors on distributed ML best practices. Suvendu holds a Master’s in Computer Science and has presented at AWS internal tech talks. He is passionate about demystifying cloud economics for ML practitioners."