MLOps Engineer

Responsibilities:

• Build and maintain ML pipelines for data processing, training, evaluation, and model deployment.
• Orchestrate batch and training jobs in Kubernetes, handling retries, failures, and resource constraints.
• Design systems that scale dynamically from small GPU jobs to thousands of GPUs on-demand.
• Collaborate with researchers to productionize their experiments into reproducible, robust workflows.
• Implement model serving endpoints (REST/gRPC) and integrate with internal tooling.
• Set up monitoring, logging, and KPI tracking for ML pipelines and compute jobs.
• Automate CI/CD and infra provisioning for ML workloads.
• Manage experiment tracking, model versioning, and metadata with tools like MLflow or W&B.
• Support model serving infrastructure that may be used by internal UIs or tools in the future.

Required Skills:

• Kubernetes: Strong experience orchestrating jobs, not just deploying services. You should be confident in managing training workloads, GPU scheduling, job retries, and Helm-based deployments.
• Python: Comfortable writing scripts and services that glue systems together. You don’t need to be a full-stack dev, but notebooks won’t cut it. Automation is the word here.
• ML Workflows: Familiarity with data preprocessing, training, evaluation, and deployment pipelines.
• Model Serving: Ability to expose models via FastAPI, TorchServe, or equivalent serving stacks.
• Linux: Strong CLI skills, you should know your way around debugging compute-heavy jobs.
• Experience with ML metadata systems (MLflow, W&B, Neptune).
• Know how to work side by side with AI assistants and agents.
• Ability to communicate and debate in English and Portuguese.

Nice-to-have skills:

• Experience with orchestration tools (Airflow, Argo Workflows, Prefect).
• Fluency in cloud environments (GCP, AWS, Azure).
• Ability to write lean and customized Dockerfiles and Helm charts that run smoothly.
• Exposure to distributed training frameworks (Ray, Horovod, Dask).
• Deep understanding of GPU scheduling and tuning in Kubernetes environments.
• Experience supporting LLM workloads or inference systems powering internal tools.

What You’ll Need to Succeed:

• Curiosity about how things fail and how to make them not.
• Strong debugging chops, especially in distributed, resource-constrained environments.
• A practical mindset, you know when to patch and when to fix.
• Ability to collaborate across ML, research, and backend teams.
• Ownership: you care about keeping systems reliable, scalable, and clean.