Self-Serve GPU Platform Engineer for Training & Inference

Perplexity; Oregon; Hybrid; USD 180,000 - 240,000; Full time

Job summary

Perplexity is seeking an infrastructure-focused platform engineer to own and evolve a self-serve GPU compute platform supporting training and inference workloads across cloud providers.

You will design and operate the GPU fleet, implement scheduling across clouds, and ensure high availability and observability, while collaborating with inference and cloud teams to align architecture and roadmaps.

Qualifications

• Extensive Kubernetes experience with operators and CRDs.
• Experience managing GPU clusters at scale (NVIDIA, CUDA).
• Elastic multi-cloud orchestration across providers.
• Strong distributed systems fundamentals: scheduling and fault tolerance.
• Proficiency in Go, Rust, or C++ for infrastructure code.
• Experience supporting both long-running training and high-availability inference.
• Ability to own problems end-to-end in unstructured environments.

Responsibilities

• Build a self-serve compute platform for training and inference workloads.
• Operate the GPU fleet across providers with provisioning and capacity planning.
• Develop scheduling and placement logic for capacity across clouds.
• Balance long-running training and production inference workloads.
• Own Kubernetes for GPU orchestration across clusters.
• Implement fault tolerance, autoscaling, and observability.
• Set technical direction and roadmaps with cross-team partners.

Skills

Deep Kubernetes experience

GPU clusters

Multi-cloud orchestration

Distributed systems fundamentals

Go/Rust/C++ for infra code

End-to-end ownership

Problem solving in ambiguity

Tools

CRDs

Kubernetes

Prometheus

Job description

Perplexity is seeking an infrastructure-focused platform engineer to own and evolve a self-serve GPU compute platform supporting training and inference workloads across cloud providers.

You will design and operate the GPU fleet, implement scheduling across clouds, and ensure high availability and observability, while collaborating with inference and cloud teams to align architecture and roadmaps.