Every customer in Hall A meets Nordlys through a scheduler. Some bring Slurm, some bring Kubernetes, and most want both on the same pod. This role makes that work without surprises.
About the role
You will join the Platform team, which builds the images, schedulers and storage paths that sit between bare metal and a customer's training job. The role is remote within the EEA, with a visit to Narvik every quarter to see the hardware your code runs on.
What you'll do
- Run and extend Slurm and Kubernetes clusters on dedicated GPU pods, including the GPU operator and topology-aware scheduling.
- Build tooling that hands a new pod to a customer in days, with their images and config preloaded.
- Improve checkpoint and data paths between compute and hall-local flash.
- Prepare the platform for B300 pods with liquid cooling from Q1 2027.
- Write clear docs and help customers' engineers when they get stuck.
What you'll bring
- Four or more years operating Slurm, Kubernetes or both in production.
- Comfort with Linux internals, containers and networking at the node level.
- Fluent Go or Python and an infrastructure-as-code habit.
- Experience with GPU workloads, NCCL or MPI is a strong plus.
- Clear written English and a preference for writing things down.