Nishant Revur Bharath Kumar

Bio:

I’m a Software Engineer II at Amazon, where I work on production visual AI, multimodal sensing, distributed systems, and Edge AI infrastructure.

My work has included Amazon Just Walk Out and newer visual reasoning systems that combine computer vision, sensor data, and on-device AI under real-world latency, reliability, memory, and compute constraints. I’ve worked on systems deployed across hundreds of production locations and large fleets of edge devices, with a focus on making AI systems reliable enough to operate continuously outside of controlled lab environments.

A major part of my work has been designing and optimizing perception pipelines for constrained hardware. This has included computer vision inference, object detection and tracking, GPU-accelerated processing, TensorRT optimization, model serving with vLLM, multimodal evaluation, workload shaping, and CPU/GPU coordination. I’ve also worked on methods for reducing compute requirements while preserving model accuracy and on system architectures that continue operating when sensor inputs are delayed, incomplete, or unavailable.

I’ve spent significant time bridging research and production engineering. This includes building evaluation infrastructure that makes offline experiments better reflect deployed behavior, validating model and system changes against production-like inputs, and working with research teams to understand whether new approaches are practical under real hardware and deployment constraints.

Earlier in my career at Amazon, I worked extensively on distributed sensor systems supporting Just Walk Out. That work involved production reliability, sensor-data availability, fault handling, large-scale migrations, performance optimization, and incident response across live systems. I have also contributed to patented work related to real-time sensor-data availability in distributed systems.

My broader interests are in physical AI, multimodal perception, Edge AI, production machine learning systems, distributed sensing, and the practical engineering required to move AI from research prototypes into dependable real-world deployments.

I hold an M.S. in Computer Science from the University of Southern California. During my graduate studies, I worked on distributed environmental sensing and co-authored a peer-reviewed ACM MobiSys publication on the USC AiR environmental sensor network.