5+ years of experience in building production systems focused on latency and failure behavior.
Proficiency in a systems programming language such as Rust, C++, or Go.
Demonstrated ownership of production systems, with the ability to discuss detailed debugging of failures.
Experience extending Kubernetes, particularly in low-reliability environments.
Familiarity with schedulers or load balancers as a primary project, not just a side task.
Responsibilities
Design and implement routing layers for globally distributed inference systems.
Manage capacity challenges and optimize resource allocation across cloud environments.
Oversee the growth and shrinkage of the fleet based on demand fluctuations.
Develop and maintain Rust services that support multi-cloud operations.
Collaborate with teams to troubleshoot and resolve production issues effectively.
Benefits
In-person work environment in San Francisco, fostering collaboration and problem-solving.
Competitive compensation package including equity options.
Full Job Description
The role
You would work on how the globally distributed inference runs: the routing layer deciding which machine handles a request, what happens when more work arrives than we have capacity for, and how the fleet grows and shrinks across clouds as demand moves. A lot of it is our own Rust service and the Kubernetes services that manage a multi-cloud fleet.
What we need to see
Around five or more years building real production systems where latency and failure behaviour mattered: serving systems, control planes, schedulers, databases, networking, trading systems. Not data pipelines or analytics.
A systems language you know well. Rust, C++ or Go is good.
Real ownership of something that ran under load, and the ability to talk through a production failure you debugged yourself, in detail.
Kubernetes at the level of someone who has extended it, particularly on low-reliability infrastructure.
Nice to have
Experience with a scheduler or load balancer as a real project rather than a side task; work with spot or preemptible capacity; any exposure to AI serving systems, which is genuinely optional. We can teach you our domain, but we can't teach distributed systems instincts.
Working here
San Francisco, in person, five days a week, because the hard problems get solved at a whiteboard. This is a senior role, with competitive compensation plus equity.
