AI Infra Engineer

Details

Perplexity is an AI-powered answer engine founded in December 2022. The company builds AI infrastructure and platforms to power large-scale training and inference workloads. This role sits at the intersection of SRE and Dev Engineering and partners closely with Infrastructure and Research teams to design, deploy, and optimize AI training and inference clusters running primarily on AWS.

Responsibilities

• Design, deploy, and maintain scalable Kubernetes clusters for AI model inference and training workloads.
• Manage and optimize Slurm-based HPC environments for distributed training of large language models.
• Develop robust APIs and orchestration systems for training pipelines and inference services.
• Implement resource scheduling and job management systems across heterogeneous compute environments.
• Benchmark system performance, diagnose bottlenecks, and implement improvements across training and inference infrastructure.
• Build monitoring, alerting, and observability solutions tailored to ML workloads running on Kubernetes and Slurm.
• Respond to system outages and collaborate across teams to maintain high uptime for critical training runs and inference services.
• Optimize cluster utilization and implement autoscaling strategies for dynamic workload demands.

Requirements

• Strong expertise in Kubernetes administration, including custom resource definitions, operators, and cluster management.
• Hands-on experience with Slurm workload management (job scheduling, resource allocation, cluster optimization).
• Experience deploying and managing distributed training systems at scale.
• Deep understanding of container orchestration and distributed systems architecture.
• High-level familiarity with LLM architecture and training processes (Multi-Head Attention, Multi/Grouped-Query, distributed training strategies).
• Experience managing GPU clusters and optimizing compute resource utilization.
• Expert-level Kubernetes administration and YAML configuration management.
• Proficiency with Slurm job scheduling, resource management, and cluster configuration.
• Python and C++ programming focused on systems and infrastructure automation.
• Hands-on experience with ML frameworks such as PyTorch in distributed training contexts.
• Strong understanding of networking, storage, and compute resource management for ML workloads.
• Experience developing APIs and managing distributed systems for both batch and real-time workloads.
• Solid debugging and monitoring skills with expertise in observability tools for containerized environments.

Preferred:

• Experience with Kubernetes operators and custom controllers for ML workloads.
• Advanced Slurm administration including multi-cluster federation and advanced scheduling policies.
• Familiarity with GPU cluster management and CUDA optimization.
• Experience with other ML frameworks like TensorFlow or distributed training libraries.
• Background in HPC environments, parallel computing, and high-performance networking.
• Knowledge of infrastructure as code (Terraform, Ansible) and GitOps practices.
• Experience with container registries, image optimization, and multi-stage builds for ML workloads.

Required experience:

• Demonstrated experience managing large-scale Kubernetes deployments in production.
• Proven track record with Slurm cluster administration and HPC workload management.
• Previous roles in SRE, DevOps, or Platform Engineering with focus on ML infrastructure.
• Experience supporting both long-running training jobs and high-availability inference services.
• Ideally 3–5 years of relevant experience in ML systems deployment with focus on cluster orchestration and resource management.

Compensation & Benefits

• Cash compensation range: $190,000 - $250,000.
• Final offers determined by experience and expertise; equity may be part of total compensation.
• Benefits include comprehensive health, dental, and vision insurance and a 401(k) plan.