• Developed automated release workflows and infrastructure for continuous deployment of AWS DLCs to multiple regions, through infrastructure as code (IaC) using AWS CDK.

• Designed a scalable multi-region canary testing and monitoring system using a microservice architecture, optimizing costs, improving scalability, and expanding testing capabilities across 30 regions.

• Developed CloudFormation Resource Provider support and API Handler contracts for Managed MLFlow on SageMaker

• Implemented a centralized SLO monitoring system to aggregate and visualize metrics across multiple regions, improving incident response times and visibility for leadership.

• One of the top contributors on the open-source AWS Deep Learning Containers project - https://github.com/aws/deep-learning-containers

• Developed DLC images for MLOps use cases across AWS services such as EC2, ECS, EKS, and SageMaker, and designed the test suite for multi-node training and inference for these images.

• Led the release of multiple Deep Learning Containers and AMIs across multiple DL Frameworks including PyTorch, TensorFlow, MXNet, etc., adding support for features like Horovod and Nvidia Transformer Engine.

Tonbo Imaging specializes in developing advanced cameras for night-vision, reconnaissance, and tracking, based on thermal imaging. Tonbo Imaging competes with global suppliers such as BAE Systems and FLIR in building vision systems for handheld, vehicle-mounted, and self-stabilized tracking cameras for military and commercial applications.

• Developed image processing algorithms and embedded firmware for a hybrid architecture on an ARM+FPGA platform. Developed a linux build system and applications for image and video processing on Intel Cyclone V devices.

• Developed software-hardware codesign for video processing and hardware peripheral control to reduce size, power and volume requirements by 60-70% compared to current platforms.

• Developed linux kernel module to speed up communication over serial interface with peripherals, which reduced communication latency by 80%.

• Developed multi-threaded raw video streaming applications using a linux kernel module to reduce frame drops to under 2%, and improve maximum video throughput to 90 FPS.