Software Development Engineer, Trainium Systems

Have you ever wondered what it takes to build the software systems that turn a rack of AI accelerators from inert silicon into a system a customer can run workloads on? Do you want deep ownership of a service that every one of those systems depends on to boot, run, and stay healthy?

Who are we?
Annapurna Labs is the part of Amazon that designs the custom silicon behind the cloud, including Trainium for AI training, Graviton CPUs, and the AWS Nitro system. Nearly every server in AWS runs hardware we designed. We build the chips, the systems, and the software that power some of the largest AI training and inference workloads in the world.

What do we do?
We build the software that provisions, operates, updates, and monitors the servers behind AWS's largest AI training workloads, built on Trainium, Graviton, and the AWS Nitro system. We give the people who run this hardware a complete way to stand it up, watch its health, deliver new software to it, and diagnose it, all through open, standard interfaces. We own that software from design through the running system.

Why it's high-impact?
The hardware is just silicon until our software makes it provisionable, observable, and serviceable. Every system runs through what we build. The service you own is on the path for all of them, and the people who run this hardware depend on it every day.

What's the challenge?
There are many ambiguous and difficult challenges in our fast-moving space. Our services require deep system and software expertise. You need to have the ability to work within a fast moving and startup-like environment in a large company. You will identify solutions, trying ideas, given space to fail and iterate to produce products that your customers love.

What you will do?
You own a service in the management plane, in provisioning, monitoring, telemetry, software delivery, or diagnostics, depending on where the team needs you. You take it from design to production, you work backward from what the operator needs, and you build on the interfaces the hardware teams expose.

Who would succeed in this role?
An engineer who likes building and operating a service in production and takes it the whole way. You read logs and reason across process and system boundaries, you ship in small verifiable steps, and you turn requirements into a service other people rely on.

Key job responsibilities

• Own the design and implementation of a large component or service in the management plane, in provisioning, monitoring, telemetry, software delivery, or diagnostics.
• Make the application-level trade-offs on your component, and seek direction on the calls that reach beyond it.
• Automate and simplify the team's development, test, and operations work, and mentor peers as you grow.
• work backward from customer requirements and deliver solutions that are simple.

A day in the life
You design and build the provisioning and management surface over standard interfaces like Redfish, and the boot flow operator tools drive. You build the offline software-update service and prove the signed bundle applies and recovers cleanly. You design each system against the interfaces the platform exposes and turn architecture decisions into APIs and tools a customer runs without you in the loop.

About the team
Our team builds the software suite for provisioning, monitoring, software delivery, and the diagnostics of Trainium systems, and we build all of it on open, standard interfaces. We serve the engineers and operators who run large-scale AI training and need to provision, monitor, update, and troubleshoot it on their own terms. We own it end to end.

We're a small team with a large charter. You own real systems on day one. There are no layers between you and the problem. You build it, you ship it, you operate it, you improve it.