What's changed: Created Professional Machine Learning Engineer Chapter 1 (Domain 1 "Low-code": BigQuery ML (classification/regression/ARIMA_PLUS/matrix factorization/boosted trees/autoencoders, features, ML.PREDICT); AutoML (data prep/labeling/Tabular Workflows/tabular-text-image-video-forecasting/debug); Model Garden ML APIs; industry-specific APIs (Document AI/Retail); RAG and Vertex AI Agent Builder).
1.2ML APIs, foundation models, and RAG
Understand building applications with ML APIs from Model Garden, using industry-specific APIs (Document AI API, Retail API, etc.), and implementing retrieval augmented generation (RAG) applications with Vertex AI Agent Builder.
You can solve many problems with pretrained APIs and foundation models without training your own. Consider "use" before "build."
1.2.1ML APIs and industry-specific APIs
Solve common tasks instantly with pretrained ML APIs from Model Garden (Vision/Speech/Natural Language/Translation, etc.). For industry needs, use industry-specific APIs like the Document AI API (extract/classify documents—invoices, contracts) or the Retail API (retail recommendations/search). Foundation models (e.g., Gemini) are also callable from Model Garden. Map "structured extraction from documents = Document AI," "retail recommendations/search = Retail API," and "general image/speech/language = Cloud ML APIs."
1.2.2RAG and Vertex AI Agent Builder
To have a foundation model answer grounded in your latest/proprietary data, use RAG (retrieval augmented generation). Retrieve relevant internal documents for the question and pass them as context to the foundation model to generate, reducing hallucinations with grounded answers. On Google Cloud, Vertex AI Agent Builder (formerly Vertex AI Search and Conversation) manages data stores, search, and building RAG apps/agents. Map "grounded generation over internal data = RAG (Vertex AI Agent Builder)" and "reflect fresh knowledge without retraining = RAG."
Common: requirement → means. E.g., "extract fields from invoices/contracts" = Document AI API; "retail recommendations/search" = Retail API; "general image labeling/OCR/translation" = ML APIs (Vision/Translation); "make a foundation model answer using internal docs" = RAG (Vertex AI Agent Builder); "reflect fresh info without retraining" = RAG.
Watch the mix-ups: (1) RAG (provide context via retrieval) vs fine-tuning (retrain the model itself) are different—RAG is often the lighter way to reflect fresh/proprietary knowledge. (2) Do not confuse general vs industry-specific APIs (document extraction = Document AI). (3) If a pretrained API suffices, no custom training needed—save cost/time.
1.2.3Section summary
- General tasks = pretrained ML APIs from Model Garden; document extraction = Document AI; retail = Retail API
- Grounded generation over internal data = RAG (Vertex AI Agent Builder); reflect fresh knowledge without retraining
- Consider "use" (APIs/foundation models) before "build" (custom training)
Sign in to track progress — Log in.
Quick check
(just a quick review)Q1. To extract structured fields (amount, date, vendor) from invoice PDFs, which is best?
Q2. To have a foundation model answer grounded in your latest internal docs without retraining, which is best?
Q3. To quickly implement product recommendations/search for a retail site, which is best?
Q4. To explore diverse pretrained/foundation models and call them from an app, which is best?
Q5. Which correctly distinguishes RAG from fine-tuning?