Full Job Description
The Role
The Cluster engineering team owns the software that turns thousands of wafers, servers, and switches into a cloud that stays up, stays busy, and stays debuggable. We stand clusters up from bare metal, schedule training and inference workloads across the fleet, keep it healthy, and make it observable to users, operators, and increasingly to AI agents. The stack is Go and Python on Kubernetes, running both on-premise deployments and our own cloud.
Responsibilities
• Declarative, CRD-driven automation of bare-metal networking, OS, and application software across clusters of Cerebras systems, servers, and switches, built to reconcile thousands of nodes
• Push-button cluster install, upgrade, and security patching with real downtime budgets, gated by canaries
• Kubernetes operators that schedule large inference workload: resource locks, priority queues, network topology, and health-aware placement
• gRPC control-plane services, authorization, admission webhooks, and quota policy for a multi-tenant fleet
• Metrics and log pipelines with purpose-built exporters for wafer-scale systems, servers (Redfish, IPMI), and network fabric (gNMI, sFlow), on Prometheus and Grafana, with SLOs and alerting
• Failure detection, HA control planes, and automated recovery, plus the CLIs, APIs, and MCP gateway that expose the fleet to users, operators, and AI agents
Skills and Qualifications
• 5+ years building and operating production distributed systems or infrastructure software
• Production-quality Go and Python
• Real Kubernetes depth: you have written or debugged controllers and operators, and you understand CRDs, reconciliation semantics, informer caches, admission webhooks, and RBAC
• Strong debugging skills across distributed systems, Linux, and networking
• Prometheus and Grafana as a practitioner: PromQL, exporter design, cardinality discipline, useful alerts
• Strong self-driving capability. This environment is large, fast-moving, and not fully documented, so we need engineers who build their own context, decide, and drive work across team boundaries. Learning speed matters more here than familiarity with our stack.
• Demonstrated adoption of AI in your engineering workflow: active use of coding agents, a view on where they help and where they mislead, and the rigor to verify what they produce.
• Nice to have: bare-metal or HPC fleet operations, scheduler internals, RDMA/RoCE and eBPF networking, Ceph or NVMe-oF, etcd and HA upgrades, inference serving stacks. ML research experience is not required.
