Google Cloud Professional Machine Learning Engineer (PMLE) Study Guide: AI and MLOps
Study for the current GCP-PMLE exam with low-code AI, data and model collaboration, training, serving, pipelines, and monitoring.
The Professional Machine Learning Engineer (PMLE) exam tests how to build, evaluate, deploy, and operate traditional ML and generative AI solutions on Google Cloud. It spans the full lifecycle: data preparation, model selection and training, inference, MLOps, governance, and monitoring.
GCP-PMLE is the shorthand used on this site; the official name is Professional Machine Learning Engineer. This guide follows the current certification page and the exam guide effective June 1, 2026. The new guide uses updated product names such as Gemini Enterprise Agent Platform. Use the version applicable to your exam date.
Exam domains and weights
| Official domain | Main focus | Weight |
|---|---|---|
| Architecting low-code AI solutions | BigQuery ML, AutoML, model/API selection | ~13% |
| Collaborating within and across teams to manage data and models | Data prep, notebooks, experiments, evaluation | ~16% |
| Scaling prototypes into ML models | Modeling, training, tuning, hardware | ~21% |
| Serving and scaling models | Batch/online inference, versions, scaling | ~20% |
| Automating and orchestrating ML pipelines | Pipelines, retraining, continuous delivery | ~18% |
| Monitoring AI solutions | Security, responsible AI, drift, production monitoring | ~13% |
How to study the six domains
Low-code AI and model selection
Compare BigQuery ML, Agent Platform AutoML, industry APIs, foundation models, and custom models by business need. For generative AI, review Model Garden, Gemini, Imagen, Veo, Agent Builder/RAG use cases, and cost, latency, and availability trade-offs.
Data, experiments, and model training
Know when to use data preparation tools such as BigQuery, Dataflow, or Spark. Understand feature management, notebook security, experiment tracking, and model evaluation. Training covers frameworks, custom training, hyperparameter tuning, fine-tuning, CPU/GPU/TPU selection, and distributed training.
Inference, pipelines, and monitoring
Compare batch and online inference. Practice model registry, canary or A/B rollout, private endpoints, and scaling. Orchestrate training and retraining with managed pipelines or tools such as Airflow, and keep preprocessing consistent between training and serving. Production monitoring includes data/concept drift, training-serving skew, model quality, and gen AI safety risks.
Common confusion: Selecting a foundation model does not finish the application. You still need retrieval/context, evaluation, safety filters, release controls, and ongoing monitoring. Inconsistent feature processing can also cause training-serving skew.
Study sequence and readiness check
Review conventional ML and data foundations, then cover model choice, training, inference, pipelines, and monitoring. In labs, record why a model fits, how you evaluate it, how you release it, and how you detect quality decline in production.
- Compare BigQuery ML, AutoML, foundation models, and custom training.
- Design data preparation, features, training, and evaluation.
- Choose inference patterns and hardware based on throughput, latency, and cost.
- Automate retraining and monitor drift, security, and responsible AI risks.
Keep practicing
Use end-to-end AI scenarios to derive data, model, deployment, and monitoring choices from business goals. Start GCP-PMLE practice.