Staff Platform Engineer - GPU Inference Control Plane

Togetherai; Greater London; On-site; GBP 120,000 - 180,000; Full time

Job summary

Together AI is seeking a software engineer to build the Kubernetes-native control plane that provisions and runs our GPU inference fleet. You will design a manifest-driven API where the inference team declares needs such as a cluster or a model deployment, and our controllers reconcile and manage lifecycle behind the scenes.

You will own reliability, performance optimizations, defragmentation, and scheduling improvements, shaping a platform that decouples developers from runtime complexity.

Qualifications

• Strong software engineering background in Go, Python, Rust.
• Experience with durable workflow orchestration tools such as Temporal or Cadence.
• Experience building control planes or orchestration systems (Kubernetes controllers/operators).
• Experience with event-driven architectures (message queues).
• Product mindset, internal platforms or APIs for other teams.

Responsibilities

• Build the provisioning state machine modelling the full lifecycle of a host and GPU stack.
• Build a self-service API and control plane for cluster provisioning and teardown.
• Automate self-healing: drain, repair or replace, reintegrate capacity.
• Own reliability: idempotency, retries, rollback, drift detection.
• Partner with ML platform teams to encode cluster constraints as abstractions.
• Engineer it like software: strong typing, tests, CI/CD for infrastructure code.

Skills

Kubernetes controllers

Go

Python

Rust

Event-driven

API design

Product mindset

Tools

Temporal

Cadence

Kafka

Job description

Together AI is seeking a software engineer to build the Kubernetes-native control plane that provisions and runs our GPU inference fleet. You will design a manifest-driven API where the inference team declares needs such as a cluster or a model deployment, and our controllers reconcile and manage lifecycle behind the scenes.

You will own reliability, performance optimizations, defragmentation, and scheduling improvements, shaping a platform that decouples developers from runtime complexity.