Professional Cloud Architect · Domain 2: Managing and provisioning a cloud solution infrastructure · 18% of the exam
Task 2.5: Configuring prebuilt solutions or APIs with Agent Platform.
Choosing between Google's AI APIs for search, conversation, vision, images, video and audio, using Gemini Enterprise agents and notebooks, and bringing Model Garden models into a solution.
Study it
Prebuilt AI APIs, Agent Search and Gemini Enterprise
Try each one before opening the answer. Every option is explained, with the Google Cloud documentation page that proves it.
Question 1 · choose 1
Refer to the Cymbal Retail case study. Cymbal needs to generate variations of a supplier's base product photo in different colors, change backgrounds and add text overlays for its catalog. Which Google Cloud capability should the architect use?
ACloud Vision API label detection and object localization on the photos
BVideo generation models on Agent Platform from a still frame
CDocument AI processors with the Enterprise Document OCR
DGemini image generation and editing models on Agent Platform
Show the answer and why
ACloud Vision API label detection and object localization on the photos
Incorrect
The Vision API analyzes images; it does not create or edit them.
BVideo generation models on Agent Platform from a still frame
Incorrect
Video models produce video clips, not edited still images for a catalog.
CDocument AI processors with the Enterprise Document OCR
Incorrect
Document AI extracts text and structure from documents; it does not change images.
DGemini image generation and editing models on Agent Platform
Correct
Gemini can generate images from a prompt and edit a provided image from a description, such as a new color, background or overlay.
The case's image generation and enhancement requirement maps to generative image models. On Agent Platform, Gemini image models both generate and edit images from instructions.
Refer to the Altostrat Media case study. Before Altostrat can summarize and extract metadata from its podcasts and interviews, it needs accurate text transcripts of the audio in several languages, with speakers distinguished. Which Google AI API should the architect use for this step?
AText-to-Speech with a multilingual neural voice for each show
BCloud Translation with a custom glossary
CNatural Language API entity analysis
DSpeech-to-Text with the Chirp 3 transcription model
Show the answer and why
AText-to-Speech with a multilingual neural voice for each show
Incorrect
Text-to-Speech goes the other way, producing audio from text.
BCloud Translation with a custom glossary
Incorrect
Translation works on text that already exists; it does not transcribe audio.
CNatural Language API entity analysis
Incorrect
Entity analysis needs text as input, so it comes after transcription.
DSpeech-to-Text with the Chirp 3 transcription model
Correct
Speech-to-Text turns audio into text, and Chirp 3 adds multilingual accuracy, speaker diarization and language detection.
Differentiating between the AI APIs starts with the input and output each one handles: audio to text is Speech-to-Text.
Refer to the Cymbal Retail case study. Cymbal wants customers to describe what they need in natural language on its website and mobile app and get highly relevant products from its large catalog, with relevance that improves from shopper behavior. Which Google Cloud product should the architect choose?
AFull-text indexes on the catalog tables in Cloud SQL
BA BigQuery ML model that classifies product categories
CCloud Translation of every product page into more languages
DAI Commerce Search with Cymbal's product catalog uploaded
Show the answer and why
AFull-text indexes on the catalog tables in Cloud SQL
Incorrect
Keyword indexes in the relational database are what Cymbal uses today, and they do not understand natural language.
BA BigQuery ML model that classifies product categories
Incorrect
Classifying products does not answer a shopper's query with ranked results.
CCloud Translation of every product page into more languages
Incorrect
Translation broadens language coverage but does nothing for search relevance.
DAI Commerce Search with Cymbal's product catalog uploaded
Correct
AI Commerce Search provides Google-quality search, browse and recommendations for ecommerce, using the uploaded catalog and user behavior.
For conversational product discovery, the prebuilt commerce search product is faster and better than building search on general-purpose databases.
A bank wants to use an open model such as Gemma for document classification. Its risk team requires that the model run inside the bank's own Google Cloud project and VPC network, and the data science team wants to fine-tune the weights. How should the architect bring the model into the solution?
ACall a partner model through Model as a Service on Agent Platform
BSelf-deploy the open model from Model Garden to an endpoint in the bank's project
CCall the Gemini API from Google AI Studio with an API key for each data science team
DUse Cloud Natural Language classification for every document
Show the answer and why
ACall a partner model through Model as a Service on Agent Platform
Incorrect
Partner MaaS models are serverless managed APIs; the bank does not run or tune their weights in its own project.
BSelf-deploy the open model from Model Garden to an endpoint in the bank's project
Correct
Self-deployed Model Garden models run in your own project and VPC network, and open weights can be tuned.
CCall the Gemini API from Google AI Studio with an API key for each data science team
Incorrect
AI Studio is a developer tool outside the bank's project controls and does not run a model in its VPC.
DUse Cloud Natural Language classification for every document
Incorrect
This is a prebuilt API, not an open model the team can tune and host itself.
Model Garden offers both managed APIs and models you deploy yourself. When the model must run in your own environment and be tuned, self-deployment is the path.
Employees want two things: AI answers to questions across Confluence, SharePoint and Jira that respect each user's existing permissions, and research notebooks where they upload PDFs and get grounded summaries of them. The company wants ready-made products rather than custom applications. Which TWO should the architect recommend? (Choose TWO.)
ADocument AI processors for every uploaded PDF
BGemini Enterprise connected to the company's data sources
CGemini Notebook Enterprise for source-grounded research
DBigQuery ML text models over exported pages
ECloud Translation of the wiki and ticket text
Show the answer and why
ADocument AI processors for every uploaded PDF
Incorrect
Document AI extracts structured data from documents; it is not a research or Q&A product.
BGemini Enterprise connected to the company's data sources
Correct
Gemini Enterprise gives employees permissions-aware search and AI assistance across connected sources such as Confluence, SharePoint and Jira.
CGemini Notebook Enterprise for source-grounded research
Correct
Gemini Notebook Enterprise lets users add sources such as PDFs to a notebook and get summaries grounded in them.
DBigQuery ML text models over exported pages
Incorrect
This would be a custom build on exported data and would lose each user's source permissions.
ECloud Translation of the wiki and ticket text
Incorrect
Translation changes the language of text; it does not answer questions.
Gemini Enterprise features bring AI agents and notebooks to employees without custom development, using connectors and permissions the company already has.