Remote job
Member of Technical Staff, Compute Platform
Job details
About this role
Role overview Build the customer-facing compute products that make distributed GPU capacity feel like one dependable cloud. Starting with managed Kubernetes on bare-metal infrastructure, you’ll also develop GPU virtual machines and capacity options, then own their reliability, service levels, adoption, and product direction. The work spans infrastructure engineering and product decisions, with direct input from customers.
Responsibilities - Bring up managed Kubernetes on bare metal, including hosted control planes and the GPU and networking components customers need. - Develop GPU virtual machines, beginning with full-node offerings and extending to smaller configurations where demand calls for them. - Create reserved, on-demand, and spot capacity products, including quotas, preemption behavior, and usage records for billing. - Design customer isolation into the platform and test it against security requirements. - Keep workloads available through node failures and loss of connectivity between a site and the central control plane. - Operate what you build: define and monitor service levels, respond when they slip, and upgrade Kubernetes and GPU drivers safely. - Make platform operations accessible through APIs, a CLI, and tools that work for both people and software agents; use customer feedback to guide product choices.
Requirements - Experience building and operating production software that people depend on, including responsibility when it fails. - Experience running production Kubernetes on bare metal, including provisioning, upgrades, and node operations. - Ability to diagnose production GPU issues such as driver faults, Xid errors, and NCCL hangs. - Working knowledge of KVM/QEMU, KubeVirt, and GPU passthrough. - Experience owning a product used by other developers, including its service levels. - Practical experience using coding agents for infrastructure work and judgment about which changes need human review.
Nice to have - Hosted control planes using Gardener or k0smotron; bare-metal provisioning with Ironic, Metal3, or Redfish. - Network fabrics such as Netris or Hedgehog, BlueField DPUs, GPU partitioning with MIG or NVLink, or quota and preemption systems such as Kueue.
Benefits and work setup - Fully remote within the United States; visa sponsorship is unavailable. - Health, dental, and vision insurance; unlimited paid time off; a $500 home office stipend; a high-spec laptop; and unlimited tokens.