Innovators wanted! Are you an entrepreneur? A builder? A dreamer? This role is part of an Amazon Special Projects team that takes the company’s Think Big leadership principle to the highest levels. We focus on creating entirely new products and services with a goal of positively impacting the lives of our customers. No industries or subject areas are out of bounds. We launched products such as Comprehend Medical, Echo Frames and Amazon Explore. If you’re interested in innovating at scale to address big challenges in the world, this is the team for you.
As the Sr. Systems Engineer, you will own the agent daemon that runs on compute nodes, the runtime sandbox that isolates workloads, and the local storage and caching layer that determines which assets are resident and ready to serve. You will build the detection logic that manages node availability, and the GPU capability reporting that tells the scheduler exactly what each node can handle. Your work directly determines whether the platform can deliver low-latency ML inference — and whether it can do so reliably at scale.
Key job responsibilities
- Design and implement the agent daemon on Linux, including GPU discovery, availability detection, and sandboxed container execution for ML inference workloads.
- Architect the local caching system — eviction policies, versioned manifests, disk budget management, and efficient update mechanisms that minimize data transfer for large model files.
- Build VRAM management and graceful workload migration logic.
- Integrate hardware attestation with the security engineer and support diverse hardware configurations and fleet onboarding automation.
About the team
Amazon Special Projects is a small, high-autonomy team building new-to-Amazon infrastructure from scratch. The systems component is where software meets hardware: you will build the on-node runtime — the agent, sandbox, GPU manager, and local storage layer — that makes a compute node reliable, secure, and performant. This is systems engineering at its most tangible — your code runs on real hardware and must be resilient to everything the real world throws at it.
