Software Engineer, GPU Infrastructure

Who are we?

Why this team?

Internal Infrastructure builds the infrastructure and tools used to train, evaluate, and serve Cohere's foundation models. You will work with AI researchers and ML engineers on stability, scalability, and observability. The work accelerates the models behind Cohere's platform, North.

You will run the shared Kubernetes fleets those workloads land on: multi-cloud GPU and TPU clusters, node health, performance, and the partnership with cloud providers and hardware vendors. The role includes a compensated 24x7 on-call rotation.

What you will do

• Deploy and operate GPU Kubernetes clusters across clouds, and bring up new hardware generations and providers.
• Work with cloud providers and hardware partners on reliability, performance, and cost, including RDMA and high-speed interconnects.
• Automate node, GPU, and fabric fault detection and remediation so most faults are fixed without a human.
• Improve fleet utilization, track cost, and cut the time it takes to get a faulty node back into service.
• Troubleshoot production failures and performance problems, and leave the system easier to operate than you found it.

We hire at more than one level. We set level based on the scope you've owned and how independently you drive work.

You may be a good fit if you have

• Run GPU Kubernetes clusters with thousands of GPUs for ML or HPC workloads.
• Debugged Linux, networking, and performance problems. Familiarity with RDMA, NCCL, or accelerator health is a plus.
• Written software in Python or Go. You would rather extend a solid open-source project than rebuild it.
• Worked with researchers, ML engineers, or another team whose goals were not identical to yours.
• Found a bottleneck, proposed a fix, and driven it in a fast-moving environment.

If this does not line up exactly with your experience, we still encourage you to apply.