Software Engineer, SI Infrastructure

What to Expect
Tesla’s AI Infrastructure team builds the software that turns GPUs and custom silicon into reliable, efficient training capacity for Autopilot, Optimus, Robotaxi, and Dojo. We own the systems layer between bare metal and researchers: cluster lifecycle, fleet automation, observability, scheduling interfaces, and the control-plane software that keeps multi-node training jobs healthy at scale.

In this role, you'll work as a software engineer in a high-performance environment, shipping production services and tooling that operate thousands of accelerated machines, not only configuring them by hand. You'll design, build, and operate software that provisions, monitors, heals, and optimizes our AI training fleets. You treat the cluster as a product: APIs, automation pipelines, reliability signals, and abstractions that make supercomputing capacity safe and fast for ML users. You’ll focus on distributed-systems and systems-software problems (hosts, accelerators, fabric, storage, schedulers), then encode fixes into software so the same failure class doesn’t require a hero page again. Scope and level scale with impact, from owning major automation surfaces to leading design across fleet-wide platforms.

What You’ll Do

• Keep the training fleet up; Drive reliability and efficiency of GPU clusters so research and production training progress is not gated on infrastructure toil or silent degradation
• Ship automation as product; Replace manual ops with reliable, testable software, APIs, job pipelines, operators/controllers, and libraries other engineers and on-call can depend on
• Own observability and remediation paths; Build metrics, logging, alerting, and automated remediation (e.g. Prometheus/Grafana-based stacks) that detect bad nodes, sick GPUs, fabric/storage faults, and scheduler failures and fence or repair safely at scale
• Raise training reliability through software; Model failure domains, reduce blast radius of HW/SW faults on long-running multi-node jobs, and improve effective capacity (less idle accelerators, fewer silent stragglers)
• Develop performance-oriented systems tooling; Build and extend tools that measure and improve throughput/stability across OS, network, storage, and runtime layers for real distributed training workloads
• Integrate with scheduling and workload platforms; Engineer against and improve HPC/AI scheduling and orchestration surfaces (e.g. Slurm/LSF and related platforms; containers/K8s where used) so capacity is allocatable, debuggable, and efficient
• Partner across the stack working with ML infrastructure users, networking, data center, and hardware/vendor partners to bring up capacity and expose clean software interfaces, not ad-hoc runbooks alone
• Participate in a 24×7 rotation for critical training infra; invest in SLIs/SLOs, runbooks-as-code, and permanent fixes that drive pages down over time

What You’ll Bring

• Strong software engineering instincts applied to infrastructure: you write maintainable automation and services, not only operate tickets
• Proficiency in Python and/or Bash (additional languages a plus); comfort building tools others will run in production
• Deep Linux experience in production (boot/image or config management, networking stack, performance debugging, package/driver realities on real hardware)
• GPU cluster operations or AI training infrastructure (NVIDIA stack, driver/CUDA ecosystem familiarity, multi-node job behavior)
• HPC schedulers and workload managers (Slurm, LSF, or similar); exposure to Kubernetes for ML/infra workloads a plus
• High-performance networking: RDMA, RoCEv2, Infini Band, congestion behavior, and host-side tuning for collective-heavy training
• Parallel / high-performance storage (Lustre, GPFS/Spectrum Scale, BeeGFS, or cloud/parallel equivalents) and its failure modes under training I/O
• Experience with configuration management / IaC-style automation at fleet scale (e.g. Ansible, or equivalent)
• Willingness to participate in a 24×7 on-call rotation for critical AI training infrastructure (cadence shared during hiring; escalation and automation are first-class goals of the role)
• Degree in CS, EE, or related field, or equivalent practical experience