GPU Platform Engineer — Self-Serve ML Compute

B Capital; United States; On-site; USD 180,000 - 230,000; Full time

Job summary

Perplexity is building a self-serve compute platform to simplify launching training jobs and inference workloads. You will own the GPU fleet and run pipelines across multiple clouds, ensuring consistent performance for researchers and engineers.

You will design scheduling, fault tolerance, and observability to keep workloads healthy, while integrating Kubernetes-based orchestration across providers. This role demands deep systems knowledge and end-to-end ownership.

Qualifications

• Deep Kubernetes experience including operators, CRDs, and multi-cluster federation.
• Experience managing GPU clusters at scale with NVIDIA hardware and networking (InfiniBand/RoCE).
• Orchestration of compute across multiple clouds (e.g., CoreWeave, AWS, GCP).
• Strong background in distributed systems: scheduling, resource allocation, fault tolerance under load.
• Proficiency writing infrastructure code in Go, Rust, or C++.
• Experience supporting both long-running training jobs and high-availability inference services.

Responsibilities

• Build a self-serve compute platform to launch training jobs and inference workloads without custom GPU provisioning.
• Operate and unify the GPU fleet across providers for consistent compute access.
• Develop scheduling and placement logic to maximize capacity use across clouds.
• Support long-running training and production inference workloads with reliable performance.
• Own Kubernetes orchestration for GPU clusters and develop operators/CRDs.
• Implement fault tolerance, autoscaling, and observability to handle node loss and capacity shifts.
• Set technical direction with teams to shape platform architecture and roadmap.

Skills

Kubernetes expertise

GPU clusters

Multi-cloud orchestration

Distributed systems

Go

Rust

C++

Tools

Prometheus

Grafana

CUDA

Job description

Perplexity is building a self-serve compute platform to simplify launching training jobs and inference workloads. You will own the GPU fleet and run pipelines across multiple clouds, ensuring consistent performance for researchers and engineers.

You will design scheduling, fault tolerance, and observability to keep workloads healthy, while integrating Kubernetes-based orchestration across providers. This role demands deep systems knowledge and end-to-end ownership.