Description

AWS runs the world's largest fleet of AI/ML accelerator servers. When a model with billions of parameters trains across a large scale of GPUs, every minute of downtime costs real progress. We are building the automation, diagnostics, and predictive intelligence that keeps this fleet running at peak. If you want to work at the intersection of hardware, software, and scale — where your code directly prevents customer-impacting failures — this is the role.

We are seeking a Systems Development Engineer to build automation software, diagnostic tooling, and fleet health infrastructure for our accelerated compute platforms. You will work across multiple teams and organizations to design scalable, reliable systems for our accelerated compute fleet.

What You Will Do

You will tackle problems no one has fully defined yet — spanning hardware, firmware, kernel, and software simultaneously. You will own systems end to end, writing code that prevents failures rather than reacts to them, and building automation that replaces manual toil with intelligent self-healing. You will work across PCIe topology, GPU diagnostics, Linux drivers, and telemetry pipelines to correlate signals and isolate faults at fleet scale. When your system catches a failing GPU before a training job crashes, that is your impact.

Why You Will Love It

Your automation runs at a large scale across servers in the cloud. When you ship, you see failure rates move within days. The team is small enough that your decisions shape the architecture, and large enough that you will always have experts to learn from across hardware, firmware, and software.

The Ideal Candidate

You know the full stack from bare-metal to userland. You debug at the intersection of components, not just within them. You build at cloud scale and care how your systems decisions impact customers. You are an excellent communicator who can drive alignment across hardware, software, and operations teams.

Key job responsibilities
Fleet Health & Predictive Infrastructure

1. Build and own the automation infrastructure for accelerator (AI/ML) fleet health at a large scale of servers, driving toward zero-touch operations that detect, diagnose, triage, and remediate faults without human intervention
2. Design and develop test frameworks, test coverage strategies, and diagnostic tooling to validate hardware functionality, detect faults, and ensure qualification coverage across the platform lifecycle.
3. Design predictive failure detection using telemetry, sensor data, error trending, and log correlation to identify degrading components before customer impact
4. Develop monitoring dashboards and alerting for real-time fleet health visibility across manufacturing, lab, and production environments
5. Define and track fleet health metrics: failure rates, mean time to detect and resolve issues, first-time fix rate, test dwell time, and predictive accuracy

Debugging & Troubleshooting

1. Debug complex system-level issues across compute, GPU, and networking in production — including Linux boot/runtime failures, PCIe, power, NIC, NVMe, and GPU subsystems on x86 and ARM
2. Perform root cause analysis correlating across firmware, kernel, driver, and physical layer; feed findings into manufacturing quality and design improvements

Systems Development & Automation

1. Design scalable test automation for hardware bring-up, regression, and qualification — reducing manufacturing test cycle times without sacrificing coverage through intelligent test sequencing and parallel execution
2. Build data pipelines correlating test results, sensor telemetry, and component-level data to identify systemic yield issues and drive upstream fixes
3. Develop and maintain Linux device drivers on ARM and x86; work with OS internals and accelerator/GPU software stacks
4. Build and manage tests covering all functional aspects of the system and CI/CD pipelines for rapid deployment to manufacturing lines and production fleet

Cross-Team Collaboration

1. Work across engineering teams and internal customers to ensure new accelerated compute hardware meets data path, control path, and onboarding requirements
2. Engage with ODMs and design partners on testability, diagnostic coverage, and automation requirements during hardware design and bring-up phases — influencing functional and performance readiness of the platform
3. Partner with datacenter operations to close the loop between field failures, manufacturing escapes, and design improvements

Operational Excellence

1. Participate in post-incident reviews, identify contributing causes and drive permanent fixes that eliminate whole classes of risk
2. Produce clear, maintainable documentation for systems, runbooks, and automation to enable others to operate and extend your work
3. Drive process improvements that increase team agility — reducing development friction, eliminating unnecessary gates, and improving delivery velocity

May require occasional (<10%) regional and international travel to Design and Manufacturing Partner sites.

A day in the life
You start the day reviewing overnight validation run results, triaging a cluster of GPU errors that correlate with a specific firmware version. Mid-morning, you push a fix to your diagnostic automation pipeline and validate it catches the failure pattern in your test environment. In the afternoon, you join a hardware bring-up call with your ODM partner to debug a PCIe link training failure on a new EVT board, walking the team through kernel logs and signal integrity data. You end the day reviewing a pull request from a teammate on a new telemetry correlation engine, and updating your manufacturing test coverage dashboard with the latest yield data.

About the team
The Hardware Engineering AI/ML UltraServer platform team is a group of engineers and technical program managers directly responsible for launching GPU-accelerated servers into the AWS fleet. Located in Seattle, Austin, and Cupertino, we collaborate with global development teams and ODM partners to deliver next-generation AI/ML infrastructure deployed in datacenters worldwide. We move fast with small, empowered teams delivering end-to-end — from server conception through fleet-scale operations.