SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
You will own the GPU compute infrastructure that powers model training, evaluation, and serving systems. This role sits within the Infrastructure team, working at the lowest layers of the stack from Kubernetes down to GPU hardware. You'll run the clusters, GPUs, and schedulers that research, product, and platform teams depend on, and manage capacity purchased from cloud providers.
Success is measured by GPU utilization efficiency, production reliability, and performance improvements to the training and serving stack.
Key responsibilities include:
- Running Kubernetes clusters for GPU training and inference workloads
- Improving GPU work scheduling and placement algorithms
- Detecting, draining, and repairing unhealthy GPU nodes
- Managing capacity and scaling with cloud providers
- Debugging failures across GPUs, Linux, storage, and networking layers
First-year milestones: By month 3, you'll handle cluster incidents independently and ship at least one scheduling or node health improvement. By month 6, you'll own one compute layer component outright (scheduling/placement, node health/repair, or capacity/regions). By month 12, you'll be responsible for a globally distributed GPU system that dynamically shifts capacity between training and inference.
You'll report to the VP of Engineering and collaborate daily with the Platform Infrastructure team, research and product teams consuming compute resources, and cloud provider/vendor partners.
REQUIREMENTS:
- Built and run GPU workloads on Kubernetes
- Can design a GPU cluster for training and online RL from scratch
- Debug confidently at OS layer and below (Linux, GPU drivers, networking, storage)
- Strong Python programming
Plus factors (not required):
- SRE or infrastructure engineering background
- Experience with Slurm, Kubernetes scheduling tools (Kueue, Karpenter), multi-cloud orchestration (SkyPilot), AWS services (EKS, SageMaker HyperPod), high-performance interconnects (EFA, InfiniBand, NCCL, Open MPI), or Terraform
- ML research experience is not required
