Microsoft Certified: Azure AI Apps and Agents Developer Associate (AI-103) (AI-103) AI-103 Questions 451–455 | Page 7/7

451

MCQmedium

You are developing an application that uses Azure OpenAI. You need to ensure that the application can authenticate to the Azure OpenAI service without storing credentials in code. What should you use?

A.Azure Active Directory (Azure AD) token-based authentication using managed identities.

B.Shared Access Signatures (SAS) generated for the Azure OpenAI resource.

C.Azure OpenAI API keys stored in Azure Key Vault and retrieved at runtime.

D.OAuth 2.0 client credentials flow with a client secret stored in Azure App Configuration.

AnswerA

Azure OpenAI supports Azure AD authentication, allowing applications to authenticate using managed identities without storing any credentials. The application can request a token from Azure AD and use it to call the service. This is the recommended approach for passwordless authentication and integrates with Azure role-based access control.

Why this answer

Azure AD authentication with managed identities allows your application to authenticate to Azure OpenAI without any stored credentials. The managed identity is assigned to the application's hosting resource, and Azure AD issues tokens that the application uses to access the service. This eliminates secret management and is the recommended practice.

Exam trap

The trap here is assuming that storing credentials in a secure store like Key Vault or App Configuration makes them passwordless; managed identities are the true passwordless solution.

Full explanation →

452

MCQmedium

You are extracting data from tables. The layout model is missing specific column headers. How can you improve the results?

A.Use a Custom Neural model

B.Increase image resolution

C.Use the Read API

D.Enable 'document-classify' feature

AnswerA

Custom Neural models learn field semantics from your labelled samples, so they can infer column headers from layout and context rather than relying on fixed positional templates. This directly addresses the missing-header constraint, where the prebuilt layout model cannot map columns without explicit header text.

Why this answer

When the layout model misses specific column headers, a Custom Neural model can be trained with labeled examples that include those headers, allowing the model to learn the specific structure and improve extraction. Custom Neural models are designed for extracting structured data from documents with varying layouts and can be tailored to recognize custom fields, including headers that the prebuilt layout model might overlook.

Exam trap

The trap is assuming that general OCR improvements (resolution) or other prebuilt features (Read API, document-classify) can solve a specific extraction gap; the key is recognizing that custom training is needed for custom headers.

How to eliminate wrong answers

Option B is wrong because increasing image resolution may improve OCR accuracy but does not address the model's inability to recognize specific column headers; it is not a targeted fix for missing headers. Option C is wrong because the Read API only extracts text and does not provide table structure or header recognition; it is a lower-level OCR service. Option D is wrong because the 'document-classify' feature is used for classifying document types, not for improving table header extraction.

Full explanation →

453

MCQeasy

When designing an agentic workflow in Azure AI Foundry, which component is used to manage the execution order and decision-making logic of multiple agents?

A.Orchestrator

B.Vector Index

C.Model Catalog

D.Prompt Template

AnswerA

The orchestrator defines the control flow that sequences agent invocations and routes decisions between them, satisfying the stem's requirement to manage execution order and decision-making logic across multiple agents. Other components handle model hosting, retrieval or tool access, but none govern inter-agent coordination.

Why this answer

The orchestrator component in Azure AI Foundry manages the execution order and decision-making logic of multiple agents. It coordinates tasks, routes requests, and determines which agent acts next based on the workflow. This is essential for building complex agentic systems.

Exam trap

AI-103 often tests the confusion between orchestration components and data storage or model management components, leading candidates to pick vector indexes or model catalogs for execution logic.

How to eliminate wrong answers

Option B is wrong because a vector index is used for storing and retrieving embeddings for similarity search, not for orchestrating agents. Option C is wrong because the model catalog is a repository of prebuilt models, not a control mechanism for agent execution. Option D is wrong because a prompt template is a reusable text structure for prompts, not a logic manager.

Full explanation →

454

MCQmedium

You are configuring an agent that must call a REST API to get current stock prices. Where must you define this integration to ensure the agent can reliably call the tool?

A.In the Azure OpenAI deployment configuration.

B.In the Search index field mapping.

C.In the Agent Tool Definition.

D.In the System Prompt.

AnswerC

Defining the REST API call in the Agent Tool Definition gives the agent the endpoint schema, parameters and authentication it needs at invocation time. This satisfies the stem's reliability constraint: the tool contract is bound to the agent, so stock-price calls resolve consistently rather than depending on external orchestration.

Why this answer

In Azure AI Foundry, tools (including REST API calls, function calling, and OpenAPI-defined tools) are declared in the Agent Tool Definition, which the agent runtime uses to decide when and how to invoke the tool. This is where you register the API schema, authentication, and parameters so the agent can reliably call the stock price endpoint.

Exam trap

AI-103 often tests the misconception that the system prompt is where tools are defined — candidates must know the System Prompt only instructs behavior, while the Agent Tool Definition provides the callable schema and integration.

How to eliminate wrong answers

Option A is wrong because the Azure OpenAI deployment configuration controls model selection, capacity, and content filtering — it does not define tool integrations for an agent. Option B is wrong because the Search index field mapping is specific to Azure AI Search and controls how fields are indexed/retrieved, not how an agent calls a REST API. Option D is wrong because the System Prompt can instruct the model to use a tool, but it does not define the tool's schema, endpoint, or authentication — without a tool definition, the model has no callable function.

Full explanation →

455

Multi-Selectmedium

You are monitoring the health and operational status of an Azure AI multi-service account. Which TWO built-in metrics can you track in Azure Monitor to evaluate performance and reliability? (Choose two.)

Select 2 answers

A.Employee payroll deductions

B.Server Errors (failed requests due to server-side issues)

C.Virtual machine disk defragmentation percentage

D.Total Calls (request volume)

E.Browser cache hit ratio

AnswersB, D

Server Errors is a built-in Azure Monitor metric counting requests that failed due to server-side faults, typically HTTP 5xx responses. It satisfies the stem's reliability requirement by surfacing backend failures, enabling alerts and diagnosis of service-side degradation on the multi-service account.

Why this answer

Option B, Server Errors, is correct because Azure Monitor exposes this metric for an Azure AI multi-service account to count failed requests caused by server-side issues, which directly reflects reliability and helps diagnose service-side faults. Option D, Total Calls, is correct because it tracks the overall request volume hitting the account, a key performance and usage indicator for capacity and trend analysis. The remaining options do not belong: employee payroll deductions (A) is an HR/finance concept unrelated to Azure Monitor metrics, virtual machine disk defragmentation percentage (C) is a VM storage metric not exposed for an AI multi-service account, and browser cache hit ratio (E) is a client-side web performance measure rather than an Azure Monitor metric for this resource.

Exam trap

AI-103 often tests whether candidates can distinguish platform metrics emitted natively by an Azure AI resource from unrelated infrastructure or business metrics, so distractors are deliberately drawn from VM, network, and HR domains.

Full explanation →

Microsoft Certified: Azure AI Apps and Agents Developer Associate (AI-103) (AI-103) (AI-103) — Questions 451–455