Founding ML Platforms Engineer: GPU Orchestration & Scale

Cumulus Labs (YC W26); San Francisco (CA); On-site; USD 180,000 - 260,000; Full time

Job summary

Cumulus Labs (YC W26) is building the systems layer for AI infrastructure in San Francisco. We seek an ML Platforms Engineer to build and run the orchestration layer that schedules workloads, allocates GPUs, and keeps a heterogeneous, multi-cloud fleet running efficiently.

You'll own the GPU orchestrator end-to-end, implement multi-tenant primitives, and drive observability with metrics, logs, and traces at scale.

Qualifications

• Strong fundamentals in data structures and algorithms, with distributed systems knowledge.
• Real production experience maintaining high-availability systems.
• Excellent design instincts and ability to trade off decisions.
• Fast learner capable of deep work; Go/Kubernetes experience is a plus, not required.
• Proficiency with modern AI coding tools (Claude Code) to move fast with rigor.
• Interested in in-person collaboration in a small team solving novel problems.

Responsibilities

• Build and extend GPU orchestrator: scheduling, fractional allocation, live workload migration across GPUs with no downtime
• Design multi-tenant primitives: quotas, isolation, usage metering, tenant-facing inference gateway
• Own observability: metrics, logs, traces at scale
• Debug hard, systems-level issues from scheduling to GPU memory and networking
• Make architectural decisions and own systems end-to-end
• Ship fast and work directly with the founder

Skills

Data structures

Algorithms

Distributed systems

Production experience

Design instincts

Go/Kubernetes familiarity

Claude Code familiarity

In-person collaboration

Tools

Go

Kubernetes

Job description

Cumulus Labs (YC W26) is building the systems layer for AI infrastructure in San Francisco. We seek an ML Platforms Engineer to build and run the orchestration layer that schedules workloads, allocates GPUs, and keeps a heterogeneous, multi-cloud fleet running efficiently.

You'll own the GPU orchestrator end-to-end, implement multi-tenant primitives, and drive observability with metrics, logs, and traces at scale.