GPU Platform Engineer – Self-Serve, Multi-Cloud Kubernetes

Perplexity; California City (CA); On-site; USD 150,000 - 210,000; Full time

Job summary

Perplexity is seeking an experienced platform engineer to own the GPU infrastructure that powers research and production workloads. You will build a self-serve compute platform and write the Kubernetes operators that run across multiple clouds.

You will design scheduling, provisioning, and capacity planning to maximize utilization while guaranteeing latency for inference services. This role demands deep systems knowledge and end-to-end ownership.

Qualifications

• Experience with managing large GPU clusters and CUDA networking.
• Expertise in Kubernetes operators, CRDs, and multi-cluster federation.
• Experience across multiple cloud providers (AWS, GCP, or similar).
• Strong distributed systems fundamentals: scheduling and fault tolerance.
• Proficient in Go, Rust, or C++ for infrastructure code.
• Ability to own problems end-to-end and operate without explicit roadmaps.

Responsibilities

• Build a self-serve compute platform for training and inference workloads.
• Operate and unify GPU fleet provisioning and capacity across providers.
• Design scheduling and placement logic for efficient resource use.
• Support long-running training and high-availability inference workloads.
• Develop Kubernetes operators and manage multiple clusters.
• Improve fault tolerance, autoscaling, and observability.
• Define platform architecture and roadmap with cross-team partners.

Skills

Deep Kubernetes experience

GPU clusters management

Multi-cloud orchestration

Distributed systems fundamentals

Go/Rust/C++ coding

SRE / reliability focus

Tools

CUDA

Triton

Prometheus

Grafana

Weights & Biases

Job description

Perplexity is seeking an experienced platform engineer to own the GPU infrastructure that powers research and production workloads. You will build a self-serve compute platform and write the Kubernetes operators that run across multiple clouds.

You will design scheduling, provisioning, and capacity planning to maximize utilization while guaranteeing latency for inference services. This role demands deep systems knowledge and end-to-end ownership.