HPC Engineer - Research Infrastructure - Remote

Overview

Luma's mission is to build multimodal AI by pushing the boundaries of what is possible with large-scale supercomputing. We are building some of the biggest and fastest AI clusters in the world, and this role is at the very heart of that effort. This requires a deep, first-principles understanding of how hardware and software intersect to unlock maximum performance.

In Short

• Architect & Optimize Supercomputers: Design, build, and tune systems that combine CPUs, GPUs (NVIDIA and AMD), and high-performance networking into world-class clusters.
• Master Low-Level Performance: Dive deep into the Linux OS, device drivers, and user-space code to optimize performance at every level of the stack.
• Debug Complex Hardware/Software Failures: Serve as the final escalation point for the most challenging GPU, networking (InfiniBand/RDMA), and system-level issues, often collaborating directly with hardware vendors like NVIDIA.
• Manage HPC Schedulers: Architect and manage modern HPC job management frameworks like Kubernetes, designing queues and partitions setups to maximize throughput and utilization for mixed research workloads.
• Build Automation for Scale: Write code to automate the monitoring, diagnostics, and healing of thousands of servers, enabling a massive infrastructure footprint with a small, elite team.

Requirements

• 8+ years of experience as an Infrastructure, DevOps, or HPC engineer working on large, complex distributed systems.
• Deep, hands-on experience managing and troubleshooting large GPU clusters from provisioning to monitoring.
• Expert in high-performance networking, with practical experience in InfiniBand, RDMA, or RoCE.
• Extensive knowledge of Linux systems, including performance tuning, debugging, and configuration.
• Deep understanding of modern HPC job management systems based on Kubernetes, and familiar with workflow orchestration frameworks like Ray or Flyte.
• Experience architecting, building, and maintaining large-scale Kubernetes clusters from first principles.
• Independently driven, tenacious problem-solver who can own issues from end-to-end.

Benefits

• Experience at national labs, research universities, or companies known for their large-scale, on-prem supercomputing infrastructure.
• Deep expertise with GPU tooling for NVIDIA and AMD GPUs, like DCGM or ROCm.