Architect low-code AI solutions
Choose a supported approach for the data and task; evaluate managed models, APIs, and foundation models against quality, cost, and latency.
Compare a warehouse classifier, a document extractor, and a grounded policy assistant. Explain the data and output requirements that drive each choice.
Go deeper: Use the official blueprint and documentation to study the named APIs, Model Garden options, and current BigQuery tuning capabilities in detail.
Collaborate on data and models
Prepare data responsibly, construct features at the right time, work securely in notebooks, and record experiments and artifacts.
Explain how late-arriving events can cause leakage and what information another team needs to reproduce an experiment.
Go deeper: Supplement these guides with current documentation and labs on Dataflow, Spark, Feature Store, Workbench, Colab Enterprise, and experiment tracking.
Scale prototypes into ML models
Select a model and training approach, tune under a budget, diagnose failures, and choose compute and parallelism.
Given low accelerator utilization, describe measurements that distinguish input, memory, and communication bottlenecks.
Go deeper: Study the product-specific training options, supported hardware, and distributed strategies in the official guide and current documentation.
Serve and scale models
Choose batch or online inference, define the serving contract, manage model versions, and design safe rollout and scaling.
Design a release with a traffic ramp, operational rollback rule, feature lookup budget, and delayed quality review.
Go deeper: Practice packaging and deploying models with the relevant Agent Platform, Cloud Run, and GKE options, including endpoint and networking limitations.
Automate and orchestrate ML pipelines
Validate data and models, preserve lineage, automate training safely, and separate retraining triggers from promotion.
Draw the data, training, evaluation, and promotion steps. Specify what happens when a segment-level gate fails.
Go deeper: Use official resources for hands-on pipeline components, orchestration choices, Cloud Build, and the exact CI/CD/continuous-training workflow.
Monitor AI solutions
Investigate skew and drift, evaluate quality across groups, secure model inputs and outputs, and monitor generative systems.
An input shifts while labels are delayed. Describe immediate checks, what remains unknown, and how to evaluate quality after labels mature.
Go deeper: Review the current monitoring and security product documentation, including supported configurations and Model Armor controls.
Practice across domains
Use the six-domain scenario set to practice making decisions, and the BigQuery time-split lab to apply data and evaluation concepts. Verify the current official guide before scheduling your exam.