MLOps Engineer

Own the JARK-Stack integration on EKS: Ray and KubeRay for distributed compute, Kubeflow Pipelines for workflow orchestration, MLflow for experiment tracking, JupyterHub for development, and advanced job schedulers (Kueue, Volcano, and Argo) for batch training. A bridge between data scientists and the platform.
Responsibilities:
Deploy and optimize Ray + KubeRay for distributed data processing and model training across GPU clusters.
Build Kubeflow Pipelines for reproducible ML workflows: data prep, training, evaluation, and deployment with lineage tracking.
Configure MLflow for centralized experiment tracking and model registry across teams.
Implement advanced job scheduling queue management, priority, preemption, and gang scheduling via Kueue/Volcano.
Build model CI/CD automated training, evaluation, validation, and canary/blue-green deployment to inference endpoints.
Create self-service tooling for data scientists' cluster provisioning, GPU allocation, and experiment templates.
Monitor ML workload performance, GPU utilization, training throughput, and data pipeline efficiency.
Requirements:
ML infrastructure / MLOps / ML platform engineering (3+ years).
Kubernetes (EKS preferred) deployments, PVs, RBAC, resource management.
At least two of: Ray/KubeRay, Kubeflow, MLflow, Airflow, Argo Workflows.
Distributed training with PyTorch DDP, Horovod, DeepSpeed, or Ray Train.
Model serving KServe, Seldon, or custom FastAPI serving.
GPU scheduling and resource management on Kubernetes.
Strong Python engineering tools and automation, not just notebooks.
Core Tech Stack: Ray/KubeRay, Kubeflow Pipelines, MLflow, JupyterHub, Argo Workflows, Kueue/Volcano, PyTorch/DeepSpeed, KServe, Helm, AWS (EKS, S3 EFS, ECR), Prometheus/Grafana.

Experience
3-7 yrs