Senior Software Engineer, GPU Cluster Infrastructure

AISafety; Berkeley (CA); Hybrid; USD 120,000 - 180,000; Full time

Benefits offered by this job

Health Insurance

401(k) match

PTO 25 days per year

Bereavement leave

WFH stipend

Catered meals (Berkeley Office)

Job summary

FAR.AI seeks an experienced infrastructure engineer to manage and scale its GPU cluster infrastructure, spanning scheduling, storage, and security. You will collaborate with researchers and other engineers to ensure high availability and efficient workload execution on large-scale GPU clusters.

The role emphasizes hands-on work with Kubernetes, batch systems, and secure multi-tenant environments, contributing to frontier AI research while maintaining robust production operations.

Qualifications

• 3+ years in systems or infrastructure engineering on production Linux
• Experience running GPU workloads with a batch layer
• Own infrastructure as code and observability for a production fleet
• Proficient programming in Python, Go, Rust, or C++
• Able to write clearly for engineers, researchers, and providers

Responsibilities

• Operate the Kubernetes GPU fleet day to day, including node lifecycle, upgrades, and capacity planning
• Own batch scheduling, quotas, priorities, preemption, and fair share across teams
• Design and run storage under the fleet with high-performance shared filesystems and object storage
• Maintain fault tolerance for multi-node training, diagnose NCCL/fabric issues, and implement checkpoint/restart patterns
• Harden the platform with identity, network policy, secrets, and sandboxing for AI agents

Skills

Kubernetes admin

GPU workloads

Infrastructure as code

Programming (Python/Go/C++)

System scalability

Tools

Terraform

Ansible

Helm

ArgoCD

Prometheus

Job description

About Us

FAR.AI is a non-profit AI research institute working to ensure advanced AI is safe and beneficial for everyone. Our mission is to facilitate breakthrough AI safety research, advance global understanding of AI risks and solutions, and foster a coordinated global response.

We’re structured to support that work from early research through real-world adoption:

• Independent by design. We can pursue what's most impactful based on our theory of change and share what we find publicly.




• A portfolio approach. Rather than focus on one single direction, we run diverse bets across the safety stack. We take promising ideas from initial experiments to deployment, informed by red-team partnerships with frontier labs and governments.




• Serious infrastructure for ambitious research. A dedicated engineering team runs our compute cluster and experiment-scaling stack, so researchers spend their time on research instead of on infra.




• Setting the standard. Our events convene key decision makers; our red-team works with frontier developers and governments; and our communications inform the public. Together, this drives adoption and sets the new standard in safety.

Since our founding in July 2022, we've grown to 50+ staff, published 40+ academic papers, and convened leading AI safety events. Our work is recognized globally, with publications at premier venues such as NeurIPS, ICML including a Best Paper Honorable Mention in 2026, and ICLR, and features in the Financial Times, Nature News, Wired Magazine and MIT Technology Review. We conduct pre-deployment testing on behalf of frontier developers such as OpenAI and independent evaluations for governments including the EU AI Office and publish the AI Security Leaderboard based on our red‑teaming expertise. We help steer and grow the AI safety field through developing research roadmaps with renowned researchers such as Yoshua Bengio; running FAR.Labs, an AI safety-focused co‑working space in Berkeley housing 40+ members; and supporting the community through targeted grants to technical researchers.

About the Team

Foundations is FAR.AI's infrastructure and engineering team. Our remit is broad: we run the compute platform, build the tools and frameworks researchers work in, automate research workflows, and help teams scale experiments well past what they'd manage alone. Our job is to accelerate the research. We do so by working directly with researchers through embedded engagements and day‑to‑day consulting, and building systems that can scale with the organization as it grows.

Foundations is growing quickly, and our infrastructure portfolio is growing fastest. We run FAR.AI's research on a mix of bare-metal and managed Kubernetes GPU clusters from multiple providers. We rent the hardware and operate the platform ourselves. The fleet has grown from dozens to hundreds of GPUs this year and it's continuing to grow quickly: we're adding providers, taking on users beyond our own researchers, and moving experiments onto frontier open‑weight models. A large amount of research is now being done by AI agents working directly on the cluster, which is driving updates to our platform infrastructure and security.

Running it well now takes dedicated specialists, so we're standing up an infrastructure sub-team that owns the cluster fleet: adding capacity, designing and managing the networking and storage under it, infrastructure as code, and the security posture, plus some of the platform layer above it. It works directly with research teams as their needs change.

About the Role

You'd work across the whole infrastructure stack, from scheduling to storage to monitoring to security, and bring real depth in at least one part of it. We're particularly interested in experience with large‑scale pre‑training and post‑training infrastructure and the network fabric under it, cluster security and sandboxing, distributed storage systems, and batch scheduling for large GPU clusters. Expertise in an adjacent area is also a good fit.

In frontier AI research, working out the infrastructure is often part of the science. You'd work directly with researchers and other engineers to keep our large‑scale experiments performant and fault‑tolerant.

We're also hiring a Tech Lead Manager, GPU Cluster Infrastructure for this team. If leading a small team while staying hands‑on sounds like you, take a look there instead.

What you'll do

• Operate the Kubernetes GPU fleet day to day. You handle node lifecycle, upgrades, driver and image rollouts, staged changes with safe rollback, and capacity planning.




• Own batch scheduling and multi‑tenancy, including queues, quotas, priorities, preemption, gang scheduling, and fair share across research teams.




• Design and run the storage under the fleet, from high‑performance shared filesystems for datasets and checkpoints to object storage tiers, quotas, and backups.




• Keep multi‑node training runs fault‑tolerant. You own node health and automated draining, debug NCCL and fabric problems, track down stragglers and flaky GPUs, and build the checkpoint and restart patterns.




• Harden the platform, covering identity and access, network policy, secrets, workload isolation, and sandboxing for the AI agents that run on the cluster.




• Bring new capacity online. You acceptance‑test providers on fabric, NCCL, and storage throughput, hold them to their SLAs, and integrate new clusters into the platform with infrastructure as code.




• Work directly with research teams on their infrastructure problems and turn the recurring ones into platform fixes. Share the on‑call rotation, runbooks, and postmortems.

Requirements

• You have 3+ years in systems or infrastructure engineering on production Linux, running GPU, HPC, or large‑scale batch platforms, and you've owned at least one system from design through operation.




• You've run production Kubernetes for GPU workloads with a batch layer on top (Slurm, Kueue, Volcano, or similar), including quotas, priority and preemption, and node health.




• You've owned infrastructure as code and observability for a production fleet, provisioning with Terraform or Ansible, deploying with Helm and ArgoCD, and monitoring with Prometheus, or their equivalents.




• You're a strong programmer in at least one language that infrastructure is commonly written in, such as Python, Go, Rust, or C++, and your automation and services are maintained as shared code.




• You write clearly for engineers, researchers, and providers, whether it's a design doc, an incident summary, or an escalation.

If you meet most of this and not all of it, we encourage you to apply anyway.

Additional skills we're excited about

Real depth in one or more of these makes you a compelling candidate:

• Distributed training infrastructure: multi‑node PyTorch and NCCL debugging, the NVIDIA node stack (drivers, GPU Operator, DCGM), InfiniBand or RoCE fabrics, topology‑aware placement.




• Distributed storage: VAST, Weka, Lustre, Ceph, or object storage at scale; checkpoint I/O.




• Cluster security: admission control, RBAC, node and container hardening, sandboxed runtimes (gVisor, Kata, Firecracker), and isolating autonomous agents on shared infrastructure.




• Scheduler internals: Kubernetes scheduler plugins or custom controllers, gang scheduling, fair‑share and quota, and the utilization, fairness, and latency tradeoffs between them.




• Multi‑provider platforms: scheduling and storage across clusters at different providers so users see one system, including clusters with no shared network and uneven data locality.

Benefits*

• Health Insurance - 94% of Insurance premium paid by Organization commencing within 1 month after your start date




• Retirement - 401(k) plan with up to 2% match




• PTO - 25 days Paid Time Off per year, accrued weekly and up to 10 days of paid sick leave per year




• Paid Leave - Paid Bereavement, Family, Medical and Pregnancy Disability Leave




• WFH Stipend & Equipment- Work computer and stipend provided for eligible employees




• Catered Meals (Berkeley Office Only) - Catered lunches and dinners on workdays at our office

*(Available only to full-time employees located in the US)

Logistics

If based in the USA or Singapore, you will be an employee of FAR.AI (501(c)(3) research non‑profit / non‑profit CLG). Outside the USA or Singapore, you will be employed via an EOR organisation on behalf of FAR.AI or as a contractor.

• Location: Both remote and in‑person (Berkeley, CA or Singapore) are possible. We sponsor visas for in‑person employees, and can hire remotely in most countries. For this role we prefer candidates whose working hours overlap with Berkeley.




• Hours: Full‑time (40 hours/week).




• On‑call: We don't run a formal on‑call rotation yet. The team is spread across time zones and covers incidents during working hours. As the experiments we run get larger we expect to introduce one, and this role would take part in it.