PMLEProfessional Machine Learning Engineer
New exam guide effective 1 Jun 2026
Design, build, productionise and monitor ML and generative AI systems on Google Cloud, from low-code solutions to custom training, serving, pipelines and monitoring.
ML engineers with about 3 years of industry experience, including at least 1 year building on Google Cloud.
- Questions
- 50–60
- Duration
- 120 min
- Cost
- USD 200
- Domains
- 6
Exam blueprint
Each domain's share of the exam, what it tests and the lessons that teach it. Hover a segment or a domain to link the two.
- Domain 1~13%
Architecting low-code AI solutions
- BigQuery ML and AutoML for tabular, forecasting and clustering problems
- Industry APIs such as Document AI, Vision and Translate
- Selecting and tuning models from Model Garden, including tuning Gemini from BigQuery
- Optimising Gemini-based applications for cost, latency and availability
Lessons for this domain
4 practice questions
- Domain 2~16%
Collaborating within and across teams to manage data and models
- Exploring and preprocessing data at scale
- Feature stores and training-serving skew
- Notebooks, experiments, metadata and lineage
- Evaluating predictive and gen AI solutions, including LLM-as-a-judge
- Handling sensitive data and PII
Lessons for this domain
5 practice questions
- Domain 3~21%
Scaling prototypes into ML models
- Choosing model families and frameworks by cost and latency
- Distributed training and accelerator selection (GPU, TPU)
- Hyperparameter tuning and regularisation
- Fine-tuning foundation models efficiently
Lessons for this domain
7 practice questions
- Domain 4~20%
Serving and scaling models
- Model registry and versioning
- Online versus batch prediction
- Autoscaling endpoints within a latency budget
- Quantisation and efficient LLM inference
Lessons for this domain
6 practice questions
- Domain 5~18%
Automating and orchestrating ML pipelines
- Managed pipelines, Airflow and Ray for orchestration
- Validating data and models; consistent preprocessing for training and serving
- Retraining policies and CI/CD/CT
Lessons for this domain
6 practice questions
- Domain 6~13%
Monitoring AI solutions
- Detecting drift, skew and bias in production
- Explainability and responsible AI practices
- Generative AI risks: prompt injection, data exfiltration, unsafe output
- Monitoring, testing and evaluating gen AI solutions
Lessons for this domain
4 practice questions
Practice
Original scenario questions written to the blueprint, not recalled exam items. Start with the free diagnostic, then drill the domains it flags.
Diagnostic
FreeA short, untimed set that samples the blueprint, with an explanation after every answer. Use it to find your weakest domain.
- Questions
- 7
- Domains covered
- 6 of 6
Domain drill
ProUp to 20 questions from a single domain, with explanations. Aim it at the weakest domain from your diagnostic.
Draws on the full 32-question bank for this exam. The diagnostic stays free.
Timed mock
ProQuestions sampled in proportion to the domain weights, under a clock scaled from the real exam. No feedback until you submit.
Draws on the full 32-question bank for this exam. The diagnostic stays free.
No attempts yet. Start with the diagnostic: your results stay in this browser and feed the readiness panel on your dashboard.