GCP Professional ML Engineer Exam Guide

Everything you need to know about the PMLE certification exam, from preparation strategies to exam day tips.

Exam Information

2 hours

Duration

60 questions

Questions

~70% (not officially published)

Passing Score

$200 USD

Exam Cost

Recommended Study Plan

Week 1

Low-Code ML Solutions

Architecting Low-Code ML Solutions

8-10 hours

BigQuery ML: CREATE MODEL, ML.PREDICT, ML.EVALUATE
Vertex AI AutoML: Tabular, Image, Text, Video
Pre-built APIs: Vision, Speech, Natural Language, Document AI, Translation
Vertex AI Agent Builder and RAG architectures
When to use BigQuery ML vs AutoML vs custom training

Recommended Resources:

Google Cloud Skills Boost

PMLE Study Notes

Week 2

Collaboration & Responsible AI

Collaborating Within and Across Teams

6-8 hours

Data validation with TensorFlow Data Validation (TFDV)
Vertex Explainable AI: SHAP, XRAI, Integrated Gradients
Model cards and documentation standards
Vertex AI Experiments and ML Metadata for lineage
Fairness metrics, What-If Tool, bias evaluation

Recommended Resources:

Responsible AI Practices

PMLE Study Notes

Week 3

Scaling Prototypes

Scaling Prototypes Into ML Models

8-10 hours

Vertex AI Custom Training: CustomJob, pre-built vs custom containers
Distributed training: MirroredStrategy, MultiWorkerMirroredStrategy, DDP
Hyperparameter tuning: Vizier, Bayesian optimization, parallel trials
GPU selection: T4, V100, A100 for training vs inference
Efficient data pipelines: tf.data, streaming from Cloud Storage/BigQuery

Recommended Resources:

Vertex AI Docs

PMLE Study Notes

Week 4

Model Serving

Serving and Scaling Models

8-10 hours

Vertex AI Endpoints: online prediction, auto-scaling, traffic splitting
Batch prediction jobs for large-scale offline inference
Model optimization: quantization, pruning, distillation, TensorRT
Containers for serving: TF Serving, TorchServe, custom containers
A/B testing and canary deployments on Vertex AI

Recommended Resources:

Vertex AI Docs

PMLE Study Notes

Week 5

ML Pipelines & MLOps

Automating and Orchestrating ML Pipelines

8-10 hours

Vertex AI Pipelines: KFP SDK, @component, @pipeline decorators
Google Cloud Pipeline Components for AutoML and Vertex AI
CI/CD for ML: Cloud Build, Cloud Deploy, Artifact Registry
Feature Store: entity types, online/offline serving
Scheduling and event-driven retraining triggers

Recommended Resources:

Vertex AI Pipelines Docs

PMLE Study Notes

Week 6

Monitoring & Review

Monitoring ML Solutions

8-10 hours

Types of drift: input, prediction, concept, label skew
Vertex AI Model Monitoring: skew/drift thresholds, alerts
Retraining strategies: schedule-based, performance-based, data-based
Cloud Monitoring and Cloud Logging for ML systems
Full review + 2 practice exams under timed conditions

Recommended Resources:

Vertex AI Monitoring Docs

CertStud Practice Exams

Exam Tips & Strategies

Scenario-Based Thinking

Most questions describe a real business scenario. Always ask: what is the constraint? (SQL team = BigQuery ML, no labels = AutoML, scaled production = custom training)

Know When Not to Train

Pre-built APIs (Vision, Speech, NL, Document AI) are frequently tested as the right answer when no training data is needed. Don't over-engineer.

Vertex AI is the Answer

Almost every GCP ML question involves Vertex AI in some way. Know Endpoint, Pipelines, Feature Store, Experiments, Model Registry, and Model Monitoring deeply.

Drift Types Are a Trap

Be precise: input drift vs concept drift vs prediction drift. The question will test whether you know which monitoring approach applies to each.

Back to Certification

GCP Professional ML Engineer Exam Guide

Everything you need to know about the PMLE certification exam, from preparation strategies to exam day tips.

Exam Information

2 hours

Duration

60 questions

Questions

~70% (not officially published)

Passing Score

$200 USD

Exam Cost

Recommended Study Plan

Week 1

Low-Code ML Solutions

Architecting Low-Code ML Solutions

8-10 hours

BigQuery ML: CREATE MODEL, ML.PREDICT, ML.EVALUATE
Vertex AI AutoML: Tabular, Image, Text, Video
Pre-built APIs: Vision, Speech, Natural Language, Document AI, Translation
Vertex AI Agent Builder and RAG architectures
When to use BigQuery ML vs AutoML vs custom training

Recommended Resources:

Google Cloud Skills Boost

PMLE Study Notes

Week 2

Collaboration & Responsible AI

Collaborating Within and Across Teams

6-8 hours

Data validation with TensorFlow Data Validation (TFDV)
Vertex Explainable AI: SHAP, XRAI, Integrated Gradients
Model cards and documentation standards
Vertex AI Experiments and ML Metadata for lineage
Fairness metrics, What-If Tool, bias evaluation

Recommended Resources:

Responsible AI Practices

PMLE Study Notes

Week 3

Scaling Prototypes

Scaling Prototypes Into ML Models

8-10 hours

Vertex AI Custom Training: CustomJob, pre-built vs custom containers
Distributed training: MirroredStrategy, MultiWorkerMirroredStrategy, DDP
Hyperparameter tuning: Vizier, Bayesian optimization, parallel trials
GPU selection: T4, V100, A100 for training vs inference
Efficient data pipelines: tf.data, streaming from Cloud Storage/BigQuery

Recommended Resources:

Vertex AI Docs

PMLE Study Notes

Week 4

Model Serving

Serving and Scaling Models

8-10 hours

Vertex AI Endpoints: online prediction, auto-scaling, traffic splitting
Batch prediction jobs for large-scale offline inference
Model optimization: quantization, pruning, distillation, TensorRT
Containers for serving: TF Serving, TorchServe, custom containers
A/B testing and canary deployments on Vertex AI

Recommended Resources:

Vertex AI Docs

PMLE Study Notes

Week 5

ML Pipelines & MLOps

Automating and Orchestrating ML Pipelines

8-10 hours

Vertex AI Pipelines: KFP SDK, @component, @pipeline decorators
Google Cloud Pipeline Components for AutoML and Vertex AI
CI/CD for ML: Cloud Build, Cloud Deploy, Artifact Registry
Feature Store: entity types, online/offline serving
Scheduling and event-driven retraining triggers

Recommended Resources:

Vertex AI Pipelines Docs

PMLE Study Notes

Week 6

Monitoring & Review

Monitoring ML Solutions

8-10 hours

Types of drift: input, prediction, concept, label skew
Vertex AI Model Monitoring: skew/drift thresholds, alerts
Retraining strategies: schedule-based, performance-based, data-based
Cloud Monitoring and Cloud Logging for ML systems
Full review + 2 practice exams under timed conditions

Recommended Resources:

Vertex AI Monitoring Docs

CertStud Practice Exams

Exam Tips & Strategies

Scenario-Based Thinking

Most questions describe a real business scenario. Always ask: what is the constraint? (SQL team = BigQuery ML, no labels = AutoML, scaled production = custom training)

Know When Not to Train

Pre-built APIs (Vision, Speech, NL, Document AI) are frequently tested as the right answer when no training data is needed. Don't over-engineer.

Vertex AI is the Answer

Almost every GCP ML question involves Vertex AI in some way. Know Endpoint, Pipelines, Feature Store, Experiments, Model Registry, and Model Monitoring deeply.

Drift Types Are a Trap

Be precise: input drift vs concept drift vs prediction drift. The question will test whether you know which monitoring approach applies to each.

Quick Feedback

GCP Professional ML Engineer Exam Guide

Exam Information

Recommended Study Plan

Exam Tips & Strategies

Scenario-Based Thinking

Know When Not to Train

Vertex AI is the Answer

Drift Types Are a Trap

GCP Professional ML Engineer Exam Guide

Exam Information

Recommended Study Plan

Exam Tips & Strategies

Scenario-Based Thinking

Know When Not to Train

Vertex AI is the Answer

Drift Types Are a Trap