Professional Machine Learning Engineer

Last checked October 22, 2026 against the official Google Cloud exam page.

1. Question 1Collaborate within and across teams to manage data and models

   Several teams reuse the same features for training and online prediction, and the values must match in both places. Which capability provides this?

   1. APer-user notebook folders that each hold a local copy of the feature CSV files for their own use
   2. BA shared spreadsheet that engineers update by hand with feature values before each training run
   3. CA managed feature store that keeps features consistent between offline training and online serving
   4. DA Pub/Sub topic that broadcasts each new feature value once to all of its subscribers in order
   Show answer and explanationHide answer

   Correct answer: C

   A managed feature store serves the same feature definitions to offline training and online prediction, which prevents mismatched values. A spreadsheet or local copies depend on people keeping them aligned, and a Pub/Sub topic delivers messages but does not manage feature consistency.

2. Question 2Scale prototypes into ML models

   A prototype trains on a 10,000-row sample in a notebook, and the production model must train on 500 million rows. Which approach scales the training best?

   1. ASplitting the data by hand into files that one engineer copies to laptops for separate training runs
   2. BRunning the same notebook on a larger single VM, resizing that VM repeatedly as the data grows
   3. CDownsampling the production training to the original 10,000 rows, to keep every run fast and cheap
   4. DCustom training on Vertex AI with distributed workers, reading data from BigQuery or Cloud Storage
   Show answer and explanationHide answer

   Correct answer: D

   Custom training on Vertex AI with distributed workers reads the full data from storage and parallelises the work, which is how a prototype becomes a scalable model. A single bigger VM hits limits, and downsampling discards most of the data. Manual splits on laptops are slow and hard to reproduce.

3. Question 3Serve and scale models

   A web application needs predictions returned in under 200 milliseconds for each request. Which Vertex AI option fits?

   1. AA BigQuery scheduled query that refreshes the predictions once every twenty-four hours for all rows
   2. BA nightly batch prediction job that writes its results to a Cloud Storage bucket for later reading
   3. CA Vertex AI Pipelines run that someone starts by hand each time a user wants a prediction
   4. DAn online prediction endpoint with autoscaling and a deployed model version serving the requests
   Show answer and explanationHide answer

   Correct answer: D

   An online endpoint with autoscaling returns predictions for individual requests with low latency. Batch jobs and scheduled queries produce results on a schedule rather than per request, and a manually started pipeline is far too slow for a live request.

4. Question 4Automate and orchestrate ML pipelines

   A team retrains its model every week with the same steps, and it needs each run's artifacts and lineage recorded. Which service orchestrates this?

   1. AVertex AI Pipelines, which runs a defined set of steps with tracked artifacts and lineage for each run
   2. BA cron job on a single VM that runs a notebook by hand each week and copies the output to a folder
   3. CCloud Storage notifications that send an email to the team whenever a new file is uploaded to the bucket
   4. DA dashboard that displays the latest model accuracy each week for the team to read in a browser
   Show answer and explanationHide answer

   Correct answer: A

   Vertex AI Pipelines orchestrates defined steps and records artifacts and lineage for every run, which suits a repeatable weekly retrain. A cron-driven notebook and notifications lack tracked lineage, and a dashboard only displays results rather than running the pipeline.

5. Question 5Architect low-code AI solutions

   A business analyst who knows SQL needs a churn model built on data that is already in BigQuery, without writing Python. Which approach is most appropriate?

   1. AA custom TensorFlow model trained on a GKE cluster from a Jupyter notebook that the analyst maintains
   2. BBigQuery ML, which creates and evaluates models with SQL statements inside the data warehouse
   3. CExporting the data to a laptop and training a scikit-learn model there, then uploading its results
   4. DA Dataflow job that writes a new feature table each time a prediction request arrives from a client
   Show answer and explanationHide answer

   Correct answer: B

   BigQuery ML trains and evaluates models with SQL where the data already lives, which matches an analyst's skills and avoids moving data. A custom TensorFlow cluster and a laptop workflow require engineering effort and copies of the data. Writing feature tables per request is an inference pipeline, not a way to build the model.