GPU ML Systems Engineer: Accelerate AI Workloads (Remote)

SkyPilot Cloud Consulting Inc.; United States; Hybrid; USD 180,000 - 240,000; Full time

Benefits offered by this job

Competitive compensation

Equity

Medical, dental, vision coverage for你

Company 401(k) plan

Autonomy and ownership

Open-source infra startup from Berkley

Gourmet lunch & dinner

Job summary

SkyPilot Cloud Consulting Inc. in San Mateo, CA is seeking an engineer to own the GPU and ML-systems layer: accelerator scheduling and utilization, the serving path, and integrations that make SkyPilot the fastest, most cost-efficient place to run demanding AI workloads.

You’ll optimize training and serving, enable large-scale pre-training with node hot-swapping, design storage for fast model checkpoints and multi-cluster serving, and deepen integrations with PyTorch, vLLM, CUDA and related

Qualifications

• Hands-on experience with GPU or accelerator systems and ML training/inference infrastructure.
• Strongly preferred: familiarity with modern ML ecosystem (vLLM, PyTorch, CUDA) and workload-orchestration frameworks (Kueue, KAI, KServe).
• Experience operating large-scale training or high-throughput inference in production.

Responsibilities

• Own GPU scheduling, utilization and health across clouds and Kubernetes.
• Build optimizations for training and serving, including storage, checkpointing, container migration, and autoscaling.

Skills

GPU/accelerator systems experience

ML training/inference infrastructure

Python

systems-level programming

workload orchestration (Kueue/KAI/Kube

PyTorch

CUDA

vLLM

Tools

Kubernetes

Kueue

KServe

KAI

PyTorch

CUDA

Job description

SkyPilot Cloud Consulting Inc. in San Mateo, CA is seeking an engineer to own the GPU and ML-systems layer: accelerator scheduling and utilization, the serving path, and integrations that make SkyPilot the fastest, most cost-efficient place to run demanding AI workloads.

You’ll optimize training and serving, enable large-scale pre-training with node hot-swapping, design storage for fast model checkpoints and multi-cluster serving, and deepen integrations with PyTorch, vLLM, CUDA and related