Google Cloud Professional Machine Learning Engineer
Validates ability to build, evaluate, productionize, and optimize AI solutions using Google Cloud capabilities and knowledge of conventional ML approaches. Covers architecting low-code AI solutions with BigQuery ML, ML APIs, and AutoML; collaborating to manage data and models with preprocessing, Jupyter notebooks, and experiment tracking; scaling prototypes into ML models including building, training, and hardware selection; serving and scaling models with batch/online inference and model registry; automating and orchestrating ML pipelines with Vertex AI Pipelines and CI/CD; and monitoring AI solutions for risks, drift, and responsible AI practices. 50-60 multiple-choice and multiple-select questions in 2 hours. Recommended 3+ years industry experience; 2-year validity.
Exam domains
- Automating and orchestrating ML pipelines22%
Developing end-to-end ML pipelines (data validation, data transformation, model training, model evaluation, model versioning, model deployment; reproducibility of training, deployment, and serving). Automating model retraining (data and concept drift detection; triggering pipelines on schedule and event; CI/CD for ML - integration with Cloud Build, GitHub Actions; integration of pipelines with Vertex AI Pipelines, Kubeflow Pipelines, TFX). Tracking and auditing metadata (Vertex AI Metadata; Vertex AI Model Registry; Vertex AI Experiments; lineage tracking; sharing and reuse of pipelines and components).
- Serving and scaling models20%
Serving models (batch and online inference - Vertex AI Predictions, Vertex AI Pipelines, Cloud Run with custom containers; using different frameworks - TensorFlow Serving, NVIDIA Triton Inference Server; model versioning and rollback - Vertex AI Model Registry; A/B testing - traffic splitting). Scaling online model serving (Vertex AI Endpoints - dedicated, shared; autoscaling - min/max replicas; using GPUs and TPUs for online prediction; Vertex AI Feature Store for low-latency feature serving; caching).
- Scaling prototypes into ML models18%
Building models (choosing ML framework and model architecture; modeling techniques given interpretability requirements). Training models (organizing training data - tabular, text, images, video, audio - on Google Cloud - Cloud Storage, BigQuery; ingestion of various file types - CSV, JSON, image files, parquet, databases - into training; training using different SDKs - Vertex AI custom training, Kubeflow Pipelines, TFX; training in distributed mode - parallel training - using multiple workers, multiple GPUs/TPUs; hyperparameter tuning - Vertex AI Vizier, AutoML tuning; troubleshooting ML model training failures). Choosing appropriate hardware for training (evaluation of compute and accelerator options - CPU, GPU, TPU; distributed training with TPUs and GPUs).
Sources
Questions are grounded in 100 references from official and authoritative materials.
- End-to-end user journeys for ML models | BigQuery | Google Cloud Documentation
- Automatic feature preprocessing | BigQuery | Google Cloud Documentation
- Train and use your own models | Vertex AI | Google Cloud Documentation
- Detect and extract text from images | Cloud Vision API | Google Cloud Documentation