Architecting Low-Code ML Solutions BigQuery ML, AutoML, pre-built APIs
Domain Weight
Architecting Low-Code ML Solutions accounts for 17% of the exam.
BigQuery ML
- CREATE MODEL directly in BigQuery using SQL — no data export required
- Supported model types: LINEAR_REG, LOGISTIC_REG, KMEANS, MATRIX_FACTORIZATION, BOOSTED_TREE_CLASSIFIER, AUTOML_CLASSIFIER, DNN_CLASSIFIER
- ML.PREDICT for batch inference on BigQuery tables
- ML.EVALUATE to compute model metrics
- Ideal when data already lives in BigQuery and team has SQL expertise
Vertex AI AutoML
- AutoML Tabular (classification, regression), Image, Text, Video
- No custom training code — upload dataset, configure target, train
- Uses Neural Architecture Search under the hood
- Export to TFLite, Edge TPU, TF SavedModel, or deploy to Vertex AI Endpoint
Pre-Built APIs (No Training)
| API | Use Case |
|---|
| Cloud Vision API | Image labeling, OCR, face detection, safe search |
| Cloud Speech-to-Text | Audio transcription (streaming + batch) |
| Document AI | Structured data extraction from documents/PDFs |
| Natural Language API | Sentiment, entities, classification, syntax |
| Translation API | Text translation across 100+ languages |
| Vertex AI Agent Builder | RAG-based search + conversational AI grounded in your data |
