Principal GPU Inference Reliability Engineer

Together Computer Inc; San Francisco (CA); On-site; USD 140,000 - 210,000; Full time

Benefits offered by this job

Health insurance

Retirement plan

Dental insurance

Vision insurance

401(k) plan

Short-term disability

Long-term disability

Life insurance

AD&D insurance

Commuting stipend

Team events

Flexible time off

Team lunches

Job summary

Together Computer Inc. is seeking a reliability-minded software engineer to own the production lifecycle of the GPU inference fleet and the Kubernetes‑native control plane that provisions clusters, models, and capacity.

You will triage InfiniBand/RDMA issues and extend automated self‑healing to reduce outages across fleet scale. You’ll own SLOs and error budgets for inference serving, drive incident response, and translate postmortems into durable improvements in controllers, automation, and

Qualifications

• Deep experience owning production reliability for large‑scale distributed systems.
• Hands-on Kubernetes controller/CRD patterns and remediation systems.
• Experience with multi‑tenant model serving and with running differentiated SLOs across model pools.

Responsibilities

• Own SLOs and error budgets for inference serving.
• Lead incident response and postmortems; drive systemic fixes into controllers and automation.
• Extend automated self‑healing; add detection/remediation plugins for new failure modes.
• Triage and fix production issues across fleet scale; translate failures into design improvements.

Skills

Kubernetes controllers

CRD patterns

Reconciliation loops

Observability focus

Event-driven design

Infra as software

SRE production reliability

Root cause analysis

Tools

InfiniBand/RDMA tooling

GPU/ML serving tooling

Monitoring stacks

Job description

Together Computer Inc. is seeking a reliability-minded software engineer to own the production lifecycle of the GPU inference fleet and the Kubernetes‑native control plane that provisions clusters, models, and capacity.

You will triage InfiniBand/RDMA issues and extend automated self‑healing to reduce outages across fleet scale. You’ll own SLOs and error budgets for inference serving, drive incident response, and translate postmortems into durable improvements in controllers, automation, and