ScenarioCert
Sign inStart free test
GCP PCA ยท Managing and provisioning a cloud solution infrastructure (17.5-17.5%)

Configuring prebuilt solutions or APIs with Agent Platform

Practice questions and explanations for the "Configuring prebuilt solutions or APIs with Agent Platform" objective of the Google Cloud Professional Cloud Architect exam. Skills measured as of 2026-10-22.

Skills in this objective

  • Differentiating between the Google AI APIs (e.g., Search, Conversation, Vision, Image, Video, and Audio)
  • Integrating Gemini Enterprise features (AI Agents and NotebookLM) to enhance workflows
  • Integrating AI models from Model Garden into the solution

Test yourself on this domain

Take the free 15-question test, or unlock all 150 questions and practice this domain on its own.

Free testUnlock for $29

Sample questions

3 of the 3 questions we have for this objective, shown with the answer and the reasoning.

Configuring prebuilt solutions or APIs with Agent Platform

The legal department at Ashgrove Pharmaceuticals reviews 200 supplier contracts each quarter. Attorneys want to upload the contracts, ask questions such as which contracts allow termination for convenience, and get answers grounded only in those documents with citations back to the source passages. The company already licenses Gemini Enterprise and wants a ready-made capability with enterprise access controls rather than a custom application.

What should you recommend?

  1. AGemini Notebook Enterprise (formerly NotebookLM Enterprise) notebooks in Gemini Enterprise, with the contracts added as sources and shared with the legal teamGemini Notebook Enterprise grounds its answers in the sources added to a notebook, cites the passages it used, and keeps the data in the organization's Google Cloud project under its identity and access controls.
  2. BA custom retrieval-augmented generation application on GKE that uses a vector database and a Gemini modelA custom RAG application could work, but building and operating it is the custom development the company wants to avoid.
  3. CCloud Vision API text detection on the contract scansText detection extracts text from images but does not answer questions about the contracts or cite passages.
  4. DCloud Translation API with a custom glossary for legal termsTranslation converts text between languages; it does not analyze contract terms or answer questions.
Why: Gemini Enterprise bundles ready-made capabilities for knowledge workers, including Gemini Notebook Enterprise, formerly NotebookLM Enterprise, for source-grounded research, and AI agents that search company data and automate multistep tasks across connected applications. The notebook's key property is that answers come from the documents added as sources and include citations, which is what a legal review workflow needs. Using the licensed product removes the need to build retrieval, grounding, and access control yourself. A custom RAG stack is the right answer only when the requirements go beyond what the packaged product supports.
Configuring prebuilt solutions or APIs with Agent Platform

Corvid Software wants to add text summarization to its support portal using an open-weight Gemma model. Policy requires that the model run on a dedicated endpoint in the company's own Google Cloud project so that the team controls the machine type and scaling. The team wants to minimize the effort of packaging and serving the model.

What should you do?

  1. ADownload the Gemma weights, build a custom serving container, and run it on a Compute Engine VM with a GPUSelf-hosting on a VM meets the control requirement, but building the container, serving stack, and scaling by hand is the effort the team wants to avoid.
  2. BOpen the Gemma model card in Model Garden and deploy it to an Agent Platform endpoint in the company's project with a selected GPU machine typeModel Garden provides deployment of supported open models to a dedicated Agent Platform endpoint in your project with prebuilt serving containers, so the team chooses the machine type and scaling without packaging the model itself.
  3. CCall a Gemini model through the Gemini API on Agent PlatformGemini models are served as Google-managed APIs; this does not use the required open-weight model on a dedicated endpoint the team controls.
  4. DUse the Cloud Natural Language API to summarize support ticketsNatural Language API offers classification, entity, and sentiment analysis, not a Gemma model on a dedicated endpoint, so it fails the stated policy.
Why: Model Garden is the catalog on Agent Platform for Google, partner, and open models, and it offers several integration paths. Some models are consumed as fully managed APIs, while open-weight models such as Gemma can be deployed with a few steps to a dedicated endpoint in your own project, using prebuilt serving containers and accelerators you select. The policy constraint, a dedicated endpoint with control over machine type and scaling, rules out managed API models, and the effort constraint rules out building the serving stack by hand. Model Garden also supports fine-tuning open models before deployment when the base model is not accurate enough.
Configuring prebuilt solutions or APIs with Agent Platform

Bluefin Broadcasting stores thousands of recorded sports programs as MP4 files in Cloud Storage. Editors want each file automatically tagged with the objects and activities that appear, along with the timestamps of every shot change, so they can find clips quickly. The team does not want to train a custom model.

Which API should you use?

  1. ACloud Vision APICloud Vision analyzes still images, so the team would have to extract frames itself and would still not get shot-change detection.
  2. BSpeech-to-Text APISpeech-to-Text transcribes the audio track; it does not identify visual objects or shot boundaries.
  3. CCloud Natural Language APINatural Language analyzes text for entities and sentiment, not the visual content of video files.
  4. DVideo Intelligence APIVideo Intelligence API analyzes stored video with pretrained models for label detection, shot change detection, and object tracking, and returns time-segmented results.
Why: Google's pretrained AI APIs are organized by the type of input they understand. Video Intelligence handles video files and returns labels, shot changes, and tracked objects with timestamps; Vision handles single images; Speech-to-Text and Text-to-Speech handle audio; and Natural Language handles text. Picking the right API first avoids building frame-extraction or custom pipelines around the wrong one. When pretrained labels are not specific enough, teams can move to custom models or multimodal Gemini models on Agent Platform.

Written from the public documentation. Not affiliated with Google Cloud. Official study guide.