Question bank
Microsoft Certified: Azure AI Apps and Agents Developer Associate
42 Microsoft Certified: Azure AI Apps and Agents Developer Associate questions with the correct answer and a written explanation, organised across 5 exam domains.
- 42 questions
- 5 domains
- Foundation level
certilumQuestion index
Jump straight to any question. Answers and explanations are inline.
Implement computer vision solutions
6 questions in this domain
- Question 1Single Choice
Which Azure resource must be provisioned before an application can use Azure AI Video Retrieval?
- AAzure AI Video Retrieval accountCorrect answer
- BAzure AI Vision (Computer Vision) resource
- CAzure AI Custom Vision project
- DAzure AI Document Intelligence resource
Why this is the answer
Video Retrieval requires its own Azure AI Video Retrieval resource. A Computer Vision key cannot authenticate Video Retrieval endpoints.
- Question 2Multiple Select
Which visual features can be requested in an Image Analysis 4.0 call?
- AdenseCaptionsCorrect answer
- BtagsCorrect answer
- CsmartCropsCorrect answer
- DCategorize
Why this is the answer
Image Analysis 4.0 supports denseCaptions, tags, and smartCrops. Categorize is a legacy feature and is not a valid visual feature in 4.0.
- Question 3Interactive
The code snippet uses the Image Analysis SDK to analyze an image. Which feature constant should replace the blank to detect objects with bounding boxes?
- AOBJECTSCorrect answer
- BTAGS
- CREAD
- DPEOPLE
Why this is the answer
The objects visual feature returns detected objects with names, confidence scores, and bounding boxes.
- Question 4Single Choice
What additional information does object detection return that image tagging does not provide?
- ABounding box coordinatesCorrect answer
- BConfidence score
- CDescriptive label
- DImage-wide category
Why this is the answer
Image tagging returns confidence-scored labels for the whole image, while object detection also returns bounding boxes that locate each detected object.
- Question 5Multiple Select
Which statements accurately describe Azure AI Vision Spatial Analysis?
- AIt must run as a Docker container.Correct answer
- BIt detects entry and line-crossing events.Correct answer
- CIt is a standard SaaS REST endpoint.
- DIt creates searchable video transcripts.
Why this is the answer
Spatial Analysis runs as a container and publishes events to endpoints such as Event Hubs or IoT Hub. It is not a SaaS REST API and does not create transcripts.
- Question 6Interactive
The code snippet shows a Video Indexer response being polled. Which value should replace the blank to indicate indexing is complete?
- AProcessedCorrect answer
- BSucceeded
- CCompleted
- DIndexed
Why this is the answer
Video Indexer polls until the state equals Processed. Values such as Succeeded or Completed are not used in the Video Indexer status payload.
Implement generative AI and agentic solutions
14 questions in this domain
- Question 7Single Choice
Which Azure service provides a unified project-based workspace for building and operationalizing generative AI applications?
- AAzure AI FoundryCorrect answer
- BAzure OpenAI Service
- CAzure Machine Learning
- DAzure AI Search
Why this is the answer
Azure AI Foundry is the unified cloud platform for designing, developing, and operationalizing generative AI applications. Azure OpenAI Service handles model inference, not project-based app development.
- Question 8Multiple Select
Which two statements accurately describe Azure AI Foundry and Azure OpenAI Service?
- AAzure AI Foundry provides the management and development plane for generative AI projects.Correct answer
- BAzure OpenAI Service is the inference plane that runs deployed OpenAI models.Correct answer
- CAzure AI Foundry makes Azure OpenAI Service unnecessary by running models directly in the hub.
- DAzure OpenAI Service hosts the Azure AI Foundry evaluation dashboard and prompt flow designer.
Why this is the answer
Azure AI Foundry is the management and development plane, while Azure OpenAI Service is the inference plane. They are used together; Foundry does not replace OpenAI, and OpenAI does not host Foundry tooling.
- Question 9Interactive
Review the code snippet. In the Azure OpenAI SDK call, what does the value 'gpt-support' represent?
- AThe deployment nameCorrect answer
- BThe model family
- CThe resource name
- DThe API version
Why this is the answer
In the Azure OpenAI SDK, the 'model' parameter receives the deployment name, not the base model name. The deployment name is defined when the model is deployed from the Azure AI Foundry model catalog.
- Question 10Single Choice
Where does the deployment name appear in an Azure OpenAI REST API chat completion call?
- AIn the path after /openai/deployments/Correct answer
- BIn the Authorization header
- CIn the request body as model_name
- DIn the query string as deployment
Why this is the answer
The Azure OpenAI REST endpoint uses /openai/deployments/{deployment_name}/chat/completions. The deployment name is part of the URL path.
- Question 11Multiple Select
Which two authentication methods are supported for Azure AI services SDK clients?
- AAPI keyCorrect answer
- BMicrosoft Entra ID tokenCorrect answer
- CAzure subscription ID
- DShared access signature
Why this is the answer
Azure AI services support API keys and Microsoft Entra ID tokens. Subscription ID is not a credential, and shared access signatures are used for storage, not for Azure AI service clients.
- Question 12Interactive
Review the code snippet. Which Azure AI service is the target of this client?
- AAzure OpenAI ServiceCorrect answer
- BAzure AI Language
- CAzure AI Vision
- DAzure AI Search
Why this is the answer
The namespace Azure.AI.OpenAI and the openai.azure.com endpoint identify Azure OpenAI Service. Other services use different namespaces and endpoints.
- Question 13Single Choice
What is the primary purpose of a vector store in a RAG solution?
- ATo store embeddings and perform similarity searchCorrect answer
- BTo cache chat responses
- CTo train a custom model
- DTo store user passwords securely
Why this is the answer
Vector stores index embeddings and support similarity search. This enables retrieval of semantically relevant document chunks for grounding.
- Question 14Multiple Select
Which two Azure data services can be used as vector stores for RAG?
- AAzure AI SearchCorrect answer
- BAzure Cosmos DB for MongoDB vCoreCorrect answer
- CAzure SQL Database
- DAzure Blob Storage
Why this is the answer
Azure AI Search and Azure Cosmos DB for MongoDB vCore support vector indexing. Azure SQL Database and Blob Storage do not provide native ANN vector indexes.
- Question 15Interactive
Review the code snippet. What does the value 5 specify in the vector search query?
- AThe number of nearest neighbor results returnedCorrect answer
- BThe number of embedding dimensions
- CThe chunk size of documents
- DThe minimum similarity threshold
Why this is the answer
In Azure AI Search vector queries, 'k' is the top-k parameter that controls how many nearest neighbor results are returned.
- Question 16Single Choice
What is the executable graph of LLM, Python, and tool nodes called in Azure AI Foundry?
- APrompt flowCorrect answer
- BBatch run
- CVariant
- DEvaluator
Why this is the answer
Prompt flow is the graph runtime in Azure AI Foundry. It connects LLM nodes, Python nodes, and tool nodes into an executable workflow.
- Question 17Multiple Select
Which two metrics are built-in quality evaluators in prompt flow?
- AGroundednessCorrect answer
- BRelevanceCorrect answer
- CLatency
- DToken count
Why this is the answer
Groundedness and relevance are built-in quality evaluators. Latency and token count are operational metrics, not quality evaluator scores.
- Question 18Interactive
Review the prompt flow definition. Where does the context input of answer_llm come from?
- AThe output of the format_prompt nodeCorrect answer
- BThe raw evaluation metric
- CThe connection secret
- DThe batch run ID
Why this is the answer
The answer_llm node declares context as ${format_prompt.output}, so it receives the output of the format_prompt node.
- Question 19Single Choice
Which set lists the four content safety categories in Azure AI Content Safety?
- AHate and fairness, sexual, violence, self-harmCorrect answer
- BToxic, abusive, offensive, dangerous
- CLow, medium, high, critical
- DPrompt, completion, context, tool
Why this is the answer
Azure AI Content Safety classifies content into Hate and Fairness, Sexual, Violence, and Self-Harm. Severity levels are Safe to High, not categories.
- Question 20Multiple Select
Which two actions are part of configuring grounding and safety for an agent?
- AConnect the agent to a trusted Azure AI Search index for groundingCorrect answer
- BEnable Prompt Shields to detect jailbreak and indirect prompt injectionCorrect answer
- CFine-tune the model to memorize current source documents
- DDisable content filters to improve response speed
Why this is the answer
Grounding uses a trusted index, and Prompt Shields add input safety. Fine-tuning is not grounding, and disabling filters is unsafe.
Implement information extraction solutions
5 questions in this domain
- Question 21Single Choice
Which Azure AI Document Intelligence model extracts fields from standard invoices and receipts without requiring training data?
- APrebuilt modelCorrect answer
- BCustom extraction model
- CCustom classification model
- DLayout model
Why this is the answer
Prebuilt models are trained on large sets of specific document types and require no training data.
- Question 22Multiple Select
Which outputs does the Azure AI Document Intelligence layout model return? Select all that apply.
- ATextCorrect answer
- BTablesCorrect answer
- CSelection marksCorrect answer
- DKey-value pairs
Why this is the answer
The layout model extracts text, tables, selection marks, and page-level details, but it does not extract key-value pairs.
- Question 23Interactive
In the provided C# code, what model identifier is passed to the document analysis operation?
- Aprebuilt-receipt
- Bprebuilt-invoiceCorrect answer
- Cprebuilt-id
- Dprebuilt-layout
Why this is the answer
The code passes the identifier prebuilt-invoice to AnalyzeDocument, so the invoice prebuilt model is selected.
- Question 24Single Choice
What is the minimum number of labeled documents required to train a custom extraction model in Azure AI Document Intelligence?
- A1
- B5Correct answer
- C10
- D50
Why this is the answer
Custom extraction models require at least five labeled documents; adding more labeled samples generally improves accuracy.
- Question 25Multiple Select
Which statements about custom classification models in Azure AI Document Intelligence are true? Select all that apply.
- AThey identify the type of an incoming document.Correct answer
- BThey route documents to the appropriate extraction model.Correct answer
- CThey extract key-value pairs from documents.
- DThey are used instead of prebuilt models.
Why this is the answer
Custom classification models identify document types and enable routing, while field extraction is performed by extraction models.
Implement text analysis solutions
5 questions in this domain
- Question 26Single Choice
Which Azure AI Language feature identifies and redacts sensitive personal data such as credit card numbers?
- APII detectionCorrect answer
- BNamed entity recognition
- CKey phrase extraction
- DEntity linking
Why this is the answer
PII detection is designed to find and redact sensitive personal data for compliance scenarios.
- Question 27Multiple Select
Which two pieces of information are returned by the language detection API when analyzing text? Select all that apply.
- AISO 639-1 language codeCorrect answer
- BConfidence scoreCorrect answer
- CTranslated text
- DSentiment label
Why this is the answer
Language detection returns an ISO 639-1 code and a confidence score; it does not translate text or return sentiment.
- Question 28Interactive
The Python code snippet contains a TODO comment for the method that returns the main talking points of a document. Which method should replace the TODO?
- Aextract_key_phrasesCorrect answer
- Banalyze_sentiment
- Crecognize_entities
- Ddetect_language
Why this is the answer
Key phrase extraction uses extract_key_phrases to identify the main talking points in a document.
- Question 29Single Choice
Which type of Azure AI Language summarization returns only original sentences selected from the source document?
- AExtractive summarizationCorrect answer
- BAbstractive summarization
- CCustom NER
- DPII detection
Why this is the answer
Extractive summarization selects the most important original sentences, while abstractive summarization generates new text.
- Question 30Multiple Select
Which two prerequisites are required to train a custom text model in Azure AI Language? Select all that apply.
- ALabeled training dataCorrect answer
- BAzure Blob Storage in the same regionCorrect answer
- CA deployed Azure OpenAI model
- DA Twitter developer account
Why this is the answer
Custom text models require labeled data and Azure Blob Storage in the same region as the Language resource.
Plan and manage an Azure AI solution
12 questions in this domain
- Question 31Single Choice
Which Azure AI service is primarily used for open-ended conversational and generative tasks?
- AAzure OpenAICorrect answer
- BAzure AI Search
- CAzure AI Document Intelligence
- DAzure AI Content Safety
Why this is the answer
Azure OpenAI provides GPT and o-series models for generative conversational tasks. Search is retrieval, Document Intelligence extracts forms, and Content Safety filters harmful content.
- Question 32Multiple Select
Which two Azure AI services are prebuilt, task-specific APIs for text and image analysis?
- AAzure AI LanguageCorrect answer
- BAzure AI VisionCorrect answer
- CAzure OpenAI
- DAzure AI Search
Why this is the answer
Azure AI Language and Azure AI Vision are prebuilt APIs for specific text and image tasks. Azure OpenAI is generative, and Azure AI Search is an information retrieval service.
- Question 33Interactive
Review the code block. Which Azure AI service is being called to summarize text?
- AAzure OpenAICorrect answer
- BAzure AI Language
- CAzure AI Search
- DAzure AI Content Safety
Why this is the answer
The code creates an AzureOpenAI client and sends a chat completion request to a GPT model, which is a generative Azure OpenAI workload.
- Question 34Single Choice
What is the main purpose of Azure AI Search in a retrieval augmented generation solution?
- AGrounding the model with relevant documentsCorrect answer
- BGenerating natural language answers
- CModerating harmful content
- DTranscribing audio input
Why this is the answer
Azure AI Search indexes documents and supports keyword, vector, and hybrid retrieval, providing the grounding data that reduces hallucination in generative AI.
- Question 35Multiple Select
Which protections are included in Azure AI Content Safety? Choose all that apply.
- AHate speech detectionCorrect answer
- BSelf-harm detectionCorrect answer
- CJailbreak risk detectionCorrect answer
- DPersonally identifiable information removal
Why this is the answer
Azure AI Content Safety detects hate, self-harm, sexual content, violence, jailbreak risks, and prompt injection. PII removal is not a Content Safety capability.
- Question 36Interactive
The code block obtains an Azure token without using a stored secret. Which identity type is being used?
- AManaged identityCorrect answer
- BAPI key
- CShared access signature
- DConnection string
Why this is the answer
The code uses ManagedIdentityCredential from the Azure Identity SDK, which obtains a token without storing any secret in application code.
- Question 37Single Choice
In Azure AI Foundry, which component is the top-level administrative container that owns shared infrastructure?
- AHubCorrect answer
- BProject
- CDeployment
- DModel
Why this is the answer
The hub is the top-level administrative container in Azure AI Foundry. It owns shared infrastructure, and all projects under it inherit those settings.
- Question 38Multiple Select
Which resources are owned at the hub level and shared by projects in Azure AI Foundry? Choose all that apply.
- AKey VaultCorrect answer
- BStorage accountCorrect answer
- CCompute instancesCorrect answer
- DProject-specific evaluation datasets
Why this is the answer
Key Vault, Storage account, and compute instances are hub-owned shared resources. Evaluation datasets are typically scoped to a project.
- Question 39Interactive
Examine the JSON. Which object represents the top-level administrative container in Azure AI Foundry?
- AhubCorrect answer
- Bproject
- Cstorage
- Dconnections
Why this is the answer
The JSON shows the hub owning storage, Key Vault, and connections, while the project references the hub by ID.
- Question 40Single Choice
Which authentication should be used for an Azure-hosted application calling Azure AI services in production?
- AManaged identityCorrect answer
- BAPI key
- CConnection string
- DSAS token
Why this is the answer
Managed identities are the safest production authentication method because Azure automatically manages the identity and no secret is stored in code.
- Question 41Multiple Select
Which three pillars form the foundational triad for securing and operating production AI services?
- AAuthenticationCorrect answer
- BPrivate networkingCorrect answer
- CMonitoringCorrect answer
- DModel training
Why this is the answer
Authentication, private networking, and monitoring form the foundational triad for securing and operating production-grade Azure AI applications.
- Question 42Interactive
What type of network access is configured in this JSON snippet?
- APrivate endpointCorrect answer
- BService endpoint
- CVPN gateway
- DPublic endpoint
Why this is the answer
The JSON disables public network access and defines a private endpoint connection, which is the Azure Private Link isolation model.
Frequently asked questions
Everything about the Microsoft Certified: Azure AI Apps and Agents Developer Associate question bank and how to use it.