Site Reliability Engineer

REALM; United States; On-site; USD 150,000 - 250,000; Full time

Benefits offered by this job

Meaningful early equity

Full benefits

Job summary

REALM is seeking an experienced SRE to embed with the teams running large-scale training and inference workloads. You’ll diagnose failures, tune jobs, build automation, and drive platform improvements.

Embedded with production GPU clusters, you’ll apply deep systems knowledge in networking, Linux, and Kubernetes to deliver reliable, scalable infrastructure. Equity and full benefits accompany this opportunity.

Qualifications

• Production experience running GPU clusters at scale.
• Strong grasp of distributed training failure modes and systems-level troubleshooting.
• Solid networking fundamentals across high-speed interconnect fabrics.
• Expert Linux - drivers, kernel tuning, container runtimes, profiling.

Responsibilities

• Embed with engineering teams running large-scale training and inference.
• Diagnose failures, tune jobs, and build automation.
• Drive platform improvements based on field learnings.

Skills

GPU clusters

Distributed training failures

Networking fundamentals

Linux kernel tuning

Software tooling

Customer briefing

Tools

Kubernetes with GPU workloads

Job description

This is an SRE role that puts you inside the environments you’re responsible for. Not monitoring dashboards from a distance - embedded with the engineering teams running large-scale training and inference, debugging failures alongside them, and owning the infrastructure that keeps their workloads healthy.

The Role

When a large distributed training run breaks, you’re the person who works out why and makes sure it can’t repeat. You’ll tune jobs, diagnose failures, build automation, and turn what you learn in the field into platform improvements.

Half the job is deeply technical. Half is knowing how to work inside someone else’s engineering organisation without friction.

Who they’re looking for

• Production experience running GPU clusters at scale
• Strong grasp of distributed training failure modes - you understand what’s happening at the systems level when a large run stalls
• Solid networking fundamentals across high-speed interconnect fabrics
• Expert Linux - drivers, kernel tuning, container runtimes, hardware-level profiling
• Kubernetes with GPU workloads in production; HPC scheduler experience equally valued
• Strong engineering skills - you build tools, not just scripts
• Comfortable briefing a customer’s leadership on something you found in their infrastructure
• Up to $250k + meaningful early equity
• Full benefits