Build machine learning solutions using Azure Databricks: Review

The MLflow lab was the whole point for me. Getting it to track runs, package the model, and register it in the Databricks registry—that's the workflow you're paying for. Everything before that felt like a standard Spark tutorial with an Azure skin. It's not bad, just familiar.

This is a corporate training course through and through. The instructor was sharp, knew their way around the portal, but you can feel the Microsoft curriculum design. It's built to get a team of ten up to speed on the 'approved' way of doing things, not to teach you the fundamentals of distributed computing. You follow the lab guide. You run the notebook. It works. The goal is to show you the platform's features, not to let you break things and learn why.

I spent years writing Java that ran on big in-house clusters. The appeal here is getting away from that infrastructure management. The course does a good job of showing that value proposition. Data prep with Spark SQL, feature engineering, distributed training, then hitting a button to serve the model as a REST API. It connects the dots. But half the time I felt like I was just following a recipe without understanding the kitchen. The focus is so heavy on the Databricks UI and helper functions that you barely touch the underlying Spark APIs.

Good if your new job hands you an Azure login and a Databricks workspace. It will save you a month of fumbling through documentation. Just don't expect it to make you a better engineer.