I've seen a lot of job postings for roles that look more like DevOps roles. Is it still possible to get hired just for building great models, or do we all need to become experts in CI/CD pipelines and infrastructure monitoring to be considered competitive?
3 answers
The era of the "Notebook Scientist" who just hands over a .ipynb file is mostly over at top companies. Recruiters are looking for the "full-stack" ML Engineer who understands the entire lifecycle. This includes data versioning with DVC, tracking experiments with MLflow, and setting up automated retraining loops. If you can't speak to how a model is monitored for "drift" once it's live, you'll struggle to pass senior-level interviews. Companies have realized that a model that stays on a laptop provides zero business value, so the plumbing is now just as important as the engine.
Which specific cloud platform—AWS, GCP, or Azure—is currently the safest bet for learning these MLOps tools?
I’ve found that smaller startups still value the pure modeling skills, but even they are moving toward managed services that require some infrastructure knowledge.
True, Pamela. Even with managed services, you still need to understand resource allocation so you don't accidentally run up a massive cloud bill during training.
Wayne, AWS is still the leader in market share with SageMaker, but the concepts are largely transferable. If you master the logic of containerization and orchestration on one, you can pick up the others quickly. Focus on the workflow logic rather than just the specific UI buttons.