The Machine Learning Professional is the step above the Databricks ML Associate. The exam is 120 minutes of multiple-choice questions, costs $200, is available online or at a test centre in English, and is valid for two years. No aids are allowed, there are no formal prerequisites, and Databricks recommends at least a year of hands-on experience with the tasks in the exam guide. Databricks publishes three domains: Model Development 44%, ML Ops 44% and Model Deployment 12%. The description mentions building scalable pipelines with SparkML, distributed training and hyperparameter tuning, advanced MLflow features, Feature Store concepts for automated feature pipelines, testing strategies, environment management with Declarative Automation Bundles, automated retraining, Lakehouse Monitoring for drift detection, custom model serving and rollout management. The ML Ops half is the reason to take it.
Testing ML code, managing environments across dev and prod, retraining on a trigger and watching for drift are the things that make or break ML in production, and few certifications cover them seriously. Getting comfortable with bundles and Lakehouse Monitoring is useful on its own, even if you never sit the exam. The model development half is where I have doubts. SparkML made sense when Databricks was mainly a Spark company and tabular models were the main workload. Today most teams I see either use scikit-learn or XGBoost on a single node, or work on deep learning and LLMs, and SparkML is something they inherit rather than choose.
An advanced exam built around it is testing a skill that is slowly losing relevance. I also think 12% for deployment is too low. The exam description talks up production-ready systems, but serving and rollout get the smallest slice by far. In my experience deployment is where most ML projects actually stall. Two smaller complaints. The official page has an empty field where the number of scored questions should be, and it carries a boilerplate line about SQL that seems copied from the data exams.
Neither matters much, but they suggest the page gets less attention than it should. And recertifying means sitting the current exam again at full price every two years, which compares badly with Microsoft's free annual renewals. Prep wise, start with the exam guide, take the Databricks Academy ML courses it lists, and practise in the Databricks Free Edition. Spend more time on MLflow, Feature Store and Lakehouse Monitoring than on memorising SparkML APIs, because that knowledge will stay useful after the exam.