Cluster Operations Engineer — MLOps & Orchestration

Comet Cloud; Boston (MA); Hybrid; USD 120,000 - 180,000; Full time

Job summary

Comet Cloud in the Boston area is hiring a Cluster Operations Engineer to own MLOps and orchestration for GPU clusters, including NVIDIA hardware and software stacks. The role blends platform engineering with customer interaction, focusing on Kubernetes/Slurm orchestration, DCGM, NCCL, and maintaining control plane health while translating workload needs into cluster configurations.

Ideal candidates have production Kubernetes/Slurm experience, model training expertise, and a hands-on approach in

Qualifications

• Experience running Kubernetes and/or Slurm in production.
• Expertise in training models, workload orchestration, and optimizing core infrastructure.
• Familiarity with NVIDIA toolchain or willingness to learn.
• Ability to monitor and operate control plane infrastructure.
• Proven ability to write actionable runbooks.

Responsibilities

• Manage cluster lifecycle and orchestration for GPUs.
• Configure Kubernetes/Slurm for enterprise workloads.
• Maintain NVIDIA software stack (DCGM, Base Command Manager, NCCL).
• Engage with customer ML teams to translate workload needs.
• Monitor control plane health and performance.

Skills

Kubernetes

Slurm

NVIDIA tools

Model training

Performance tuning

Customer support

Tools

DCGM

Base Command Manager

NCCL

Infiniband

Job description

Cluster Operations Engineer — MLOps & Orchestration (preferably Boston/Dallas; hybrid or remote)

After we deploy a GPU cluster, someone has to make it a great place to actually train models. That's this role.

We build dedicated NVIDIA clusters — GB300/GB200 NVL72, HGX-class B300 nodes, Quantum-3 XDR Infiniband and more for enterprise customers who get a named engineer, not a ticket queue. You'd be that engineer for the platform layer: Kubernetes and Slurm orchestration, the NVIDIA software stack (DCGM, Base Command Manager, NCCL), control plane health, and the direct customer relationship that turns their workload requirements into cluster configuration.

You're a fit if you:

• - Have run Kubernetes and/or Slurm in production and actually enjoy scheduler configuration
• - Have a expertise in training models, workload orchestration, optimizing core infrastructure, and resolving performance issues
• - Know the NVIDIA toolchain, or know one part deeply and want the rest
• - Can monitor and operate control plane infrastructure without drama
• - Like sitting with a customer's ML team as much as sitting in a terminal
• - Write runbooks people actually use

Why here: Lean team, no process layer between you and the machines, and the newest NVIDIA hardware in production. Your customers know your name.

Boston metro or Dallas | Hybrid or remote | Periodic site travel

Comet Compute is an NVIDIA Inception member building to the NVIDIA Cloud Partner reference architecture. Equal opportunity employer.