Familiarity with AMD Instinct platforms, MI300X, MI325X, MI355X, and the ROCm software ecosystem, or transferable depth with NVIDIA/CUDA, Experience diagnosing high-speed networking issues, RDMA, link performance, and the class of problems where the GPUs look healthy but throughput has collapsed. Triage GPU and node hardware faults, including nodes presenting fewer than 8 GPUs, RMA candidates, and thermal or power events, and coordinate with data center operations on remediation.