Description
You will be the first member of the new Platform Team, focused on GPU workloads and high-performance model serving & training technologies.
You will bring your DevOps expertise to help Data Scientists and Machine Learning Engineers build and operate AI services.
You will play a key role in building and operating a new Kubernetes cluster, deploying self-service applications for other teams, establishing best practices and new workflows and mentoring future hires.
Responsibilities:
- Design, build and operate a GPU platform on GCP for AI workloads, fully expressed as infrastructure-as-code
- Deliver self-service compute for the data science team, e.g. Ray on Kubernetes, with job submission, queuing, quotas and cost visibility
- Establish golden paths for deploying services so product teams can ship a new service without having to perform operations each time
- Assist the existing IT Squad in ensuring the deployment, hosting, and smooth operation of models in production environments
- Oversee and control infrastructure costs, including GPUs, cloud resources, and related services
- Collaborate closely with Data Science teams, assisting them from the PoC stage to service deployment on production clusters
- Continuously improve the performance, reliability, and efficiency of above systems
- Automate internal workflows, utilizing AI-powered systems wherever applicable
QUALIFICATIONS:
- You have significant experience as a Platform, DevOps, or MLOps Engineer, with strong Software Engineering fundamentals and a good understanding of Distributed Systems
- You have strong hands-on experience with Kubernetes, Terraform, and GitOps practices, including CRDs and operators, scheduling and autoscaling, node pool design, state management, ArgoCD, and Helm
- You have solid experience with GCP; real-world exposure to AWS is a strong plus
- You have a platform-as-a-product mindset and are comfortable working with ambiguity and autonomy: you treat data scientists as users, document your solutions, measure adoption, research solutions independently, and set technical precedents when needed
- You are fluent in English, both written and spoken, and can communicate effectively with a wide range of stakeholders
Nice to Have:
- Hands-on experience running or optimizing GPU-based workloads in production, ideally for ML training or inference
- Experience with distributed job scheduling, such as KubeRay, Slurm, or Kubeflow
- Experience with cross-cloud networking, including Interconnect, VPN, Private Service Connect, and identity federation between AWS and GCP
- Familiarity with MLOps tooling, such as MLflow, Weights & Biases, and model registries