5-7 years of experience in Linux system administration and automation
Proficient in scripting languages like Bash and Python
Hands-on experience with Docker and orchestration tools such as Kubernetes
Background in running production services with monitoring and alerting systems
Skilled in building reliability tooling for clusters
Experience managing developer tools and third-party services
Strong debugging skills across compute, storage, and networking
Responsibilities
Administer and automate Linux-based cluster environments
Onboard and offboard users, manage access control, and conduct security audits
Monitor system resources, schedule jobs, and ensure cluster health
Bring up and validate new hardware for compute, networking, and storage
Deploy and scale inference services across various accelerators
Manage CI/CD pipelines for model deployments and ensure service reliability
Plan capacity, autoscale resources, and track utilization and costs
Benefits
Comprehensive medical, dental, vision, and FSA plans
Relocation and immigration support on a case-by-case basis
In-office snacks and meals provided
Unlimited PTO and company holidays
Collaborative, high-energy work environment in San Francisco
Full Job Description
The Role:
As a Machine Learning Systems Administrator / DevOps Engineer, you will maintain and build the core infrastructure behind both our model training and our inference cloud. You'll split your time between our large-scale training clusters and the deployments and services that run our models in production on heterogeneous compute, which spans multiple accelerator types and vendors. You'll work closely with our training, inference, and product teams to keep these systems reliable, secure, and fast. You'll also lay the groundwork for workflows that can scale with us.
You'll Work Across:
Training clusters
Administration and automation of our Linux-based cluster environments
User onboarding/offboarding, access control, and security auditing
Monitoring system resources, job scheduling, and cluster health
Bring-up and validation of new hardware (compute, networking, storage)
Supporting and improving developer workflows (e.g., VSCode remote development, Docker)
Enabling and supporting large-scale training jobs
Inference cloud
Deploying, operating, and scaling inference services across heterogeneous accelerators
Managing container orchestration, CI/CD, and rollout pipelines for model deployments
Building observability for production services, including metrics, logging, alerting, and dashboards
Owning service reliability: uptime, incident response, and postmortems
Capacity planning, autoscaling, and utilization/cost tracking across hardware types
Securing production systems through secrets management, network policy, and tenant isolation
You'll own critical systems end to end. Your work will directly affect developer productivity, training throughput, and the performance and reliability of our inference platform.
What We're Looking For / Requirements:
Experience with Linux system administration, user and access management, and automation
Demonstrated expertise in scripting languages for system tooling and automation (bash, Python, etc.)
Experience with containers (Docker) and scheduling/orchestration systems (Slurm, Kubernetes)
Experience running production services, including monitoring and alerting (e.g., Prometheus, Grafana), deployment pipelines, and incident response
Experience building tooling for cluster validation and reliability (accelerator, networking, and storage health checks)
Experience setting up and managing developer tools and third-party services (e.g., cloud storage providers, Docker Hub, Slack, Gmail, Telegraf, experiment trackers)
Excellent debugging and troubleshooting skills across compute, storage, and networking
Strong communication skills and ability to collaborate across technical and non-technical teams
Qualifications / Additional Skills:
Experience with infrastructure as code (e.g., Ansible, Terraform)
Prior work supporting ML/AI infrastructure, including GPU management and workload optimization
