Staff ML Engineer – AWS Trainium & SageMaker

Role overview Lead the engineering of production model training workloads on managed cloud infrastructure using custom AI accelerators. The work spans PyTorch code, accelerator compilation and behavior, and end-to-end pipeline delivery, with a focus on diagnosing hardware-specific issues and balancing throughput with cost.

Responsibilities - Train and operate models on managed training infrastructure backed by custom accelerators. - Optimize PyTorch training code, accounting for device architecture, compiler behavior, memory use, and throughput. - Identify whether training failures stem from data, code, compilation, or accelerator behavior. - Turn training requirements into working, cost-aware production pipelines. - Tune distributed training workloads and collaborate with client and engineering teams on delivery.

Requirements - Strong practical experience with PyTorch; distributed or multi-device training is useful. - Production experience with managed model training or inference services. - Ability to debug issues close to the hardware and understand device-specific compilation. - Solid Python skills and comfort delivering production systems in a client-facing environment.

Nice to have - Experience with custom AI accelerators or their software development kits; related experience on alternative accelerator hardware is relevant.