AI-103 Practice Test 31 — 15 Questions

Question 1

Which THREE features are provided by the prebuilt Layout model in Azure AI Document Intelligence? (Choose THREE)

Accepted Answer

Extraction of text lines and words with bounding polygons. Option B is correct because the prebuilt Layout model in Azure AI Document Intelligence returns OCR results as text lines and individual words, each accompanied by bounding polygons that give the precise coordinates of the text on the page. Option C is correct because the Layout model extracts table structures, including the rows, columns, and cell spans, so tabular data is returned in an organized, machine-readable form. Option E is correct because the Layout model detects selection marks such as checkboxes and radio buttons and reports their state as selected or unselected. Option A does not belong because sentiment analysis is a natural language processing capability (for example, from Azure AI Language), not a feature of the Document Intelligence Layout model. Option D does not belong because translation is handled by Azure AI Translator, whereas the Layout model only extracts and structures document content without translating it.

Answer

Automatic sentiment analysis of paragraph text

Answer

Language translation of extracted text into English

Question 2

When building a custom template model in Azure AI Document Intelligence Studio, which THREE types of data or attributes can you extract by labeling your training documents? (Choose THREE)

Accepted Answer

Tables (including row and column structures). Option C is correct because custom template models in Azure AI Document Intelligence Studio support labeling and extracting tabular data, including the row and column structure of tables, which the service returns as structured cells with row/column indices. Option D is correct because selection marks such as checkboxes and radio buttons are a first-class labeled entity type, returned with a state of selected or unselected and a confidence score. Option E is correct because the core purpose of a custom template model is to label and extract fields, i.e., key-value pairs such as invoice number, date, or total, which are returned as named field/value pairs. Option A does not belong because facial biometric embeddings are produced by Azure AI Face, not by Document Intelligence, which analyzes document text, layout, and structure rather than identifying people. Option B does not belong because real-time audio transcription timestamps come from Azure AI Speech (for example, the Speech-to-Text API), whereas Document Intelligence processes documents and images, not live audio streams.

Answer

Facial biometric embeddings

Answer

Real-time audio transcription timestamps

Question 3

You are troubleshooting a deployment failure where an ARM template fails to provision an Azure OpenAI account. Which TWO potential causes should you investigate? Each correct answer presents part of the solution.

Accepted Answer

The subscription has exceeded its regional quota or model deployment quota limit.. Option A is correct because Azure OpenAI provisioning fails when the target subscription has hit its regional quota or per-model deployment quota limit, so you must verify available quota in the chosen region before redeploying. Option D is correct because Azure OpenAI model deployments are only offered in specific regions, and requesting a region that does not support the model causes the ARM template to fail. Options B, C, and E do not belong: Azure DevOps IP whitelisting, browser cache corruption, and uncommitted local git changes affect pipeline connectivity, client rendering, and source control state respectively, none of which block ARM template provisioning of an Azure OpenAI account.

Answer

The local workstation IP address is not whitelisted in Azure DevOps.

Answer

The client web browser cache is corrupted.

Answer

The local git repository has uncommitted changes.

Question 4

You are deploying an LLM in Azure AI Foundry and want to restrict access to the endpoint to a specific virtual network. What should you configure?

Accepted Answer

Private Endpoint. A private endpoint creates a private IP address for the Azure AI Foundry endpoint within your virtual network, ensuring traffic never traverses the public internet. This restricts access to only resources within that VNet. It is the standard Azure PaaS networking feature for private connectivity.

Answer

Azure AD Conditional Access

Answer

Firewall Rule on the Model

Answer

Managed Identity

Question 5

You are troubleshooting a custom text classification model that exhibits high training loss and low validation accuracy, indicating severe model underfitting. What adjustment should you make to improve the model?

Accepted Answer

Increase the diversity and volume of training data and ensure model training is allowed to complete fully.. Underfitting means the model has not learned the underlying patterns in the data — training loss stays high because the model cannot even fit the training set. The remedy is to give the model more signal (more diverse, higher-volume data) and to let training run to completion so it has enough iterations to converge. Lowering thresholds or shrinking data would worsen the problem, not fix it.

Answer

Lower the confidence score threshold to 0.01.

Answer

Reduce the training dataset size by 90%.

Answer

Switch the resource pricing tier from Standard to Free.

Question 6

You are designing a generative AI application that processes user uploaded documents. You must ensure that Personally Identifiable Information (PII) such as Social Security numbers are automatically redacted before the text is sent to the Azure OpenAI model. Which service should you integrate into your pipeline?

Accepted Answer

Azure AI Language service (PII detection and redaction). Azure AI Language provides a prebuilt PII detection and redaction capability that identifies entities like Social Security numbers, credit card numbers, and names, and can mask or redact them in text. Integrating it into the pipeline before sending text to Azure OpenAI ensures sensitive data never reaches the model. This is the purpose-built service for the requirement.

Answer

Azure AI Translator

Answer

Azure AI Vision OCR

Answer

Azure Monitor Application Insights

Question 7

Your development team wants to deploy a new version of an Azure OpenAI model while ensuring zero downtime for client applications currently calling the existing model deployment. What is the recommended deployment strategy?

Accepted Answer

Create a new deployment with a distinct deployment name for the new model version, then update application configuration to point to the new deployment.. Azure OpenAI supports multiple simultaneous deployments of the same or different models under distinct deployment names within a single resource. By creating a new deployment for the updated model version and then switching application configuration to the new deployment name, traffic continues to be served by the old deployment until the cutover, achieving zero downtime. This also enables easy rollback by reverting the configuration.

Answer

Overwrite the existing model deployment name with the new model version.

Answer

Delete the Azure OpenAI resource and provision a new one in the same region.

Answer

Downgrade the pricing tier to pause traffic while updating the model version.

Question 8

You are preparing training data for a custom named entity recognition (NER) project in Language Studio. Which THREE guidelines should you follow to ensure high model accuracy? (Choose three.)

Accepted Answer

Ensure consistent and accurate tagging of entities across all training documents.. Option A is correct because consistent and accurate tagging is fundamental to supervised NER training: if the same entity type is labeled inconsistently across documents, the model receives contradictory signals and cannot learn reliable entity boundaries or types. Option C is correct because Azure AI Language's custom NER guidance recommends providing sufficient labeled examples per entity type (typically at least 10-15 examples per entity, and more for complex or ambiguous entities) so the model can generalize the entity pattern rather than memorize a few instances. Option D is correct because training data should reflect the real-world distribution and variety of documents the model will encounter, including different formats, domains, and contexts, which improves generalization and reduces overfitting to a narrow sample. Option B is wrong because a single training document cannot provide enough examples or variability for the model to learn robust entity patterns, and the service requires a minimum number of documents for training. Option E is wrong because tagging every word as an entity defeats the purpose of NER, which must distinguish entities from non-entity tokens; the model also needs negative examples (words not tagged as entities) to learn what is not an entity.

Answer

Use exactly one training document for the entire project.

Answer

Tag every single word in every document as an entity.

Question 9

You are implementing an agentic workflow where an agent executes tools that modify corporate data. To adhere to Responsible AI principles, you must ensure a human reviewer approves sensitive tool actions before they are executed. What design pattern should you implement in Azure AI Agent Service?

Accepted Answer

Human-in-the-loop (HITL) approval workflow for tool execution. Human-in-the-loop (HITL) approval inserts a human review step before sensitive tool actions execute, satisfying Responsible AI requirements for oversight of consequential operations. In Azure AI Agent Service, this is implemented by pausing the agent's tool call, surfacing it for approval (via UI, API, or Logic Apps), and only proceeding on explicit human consent. This preserves accountability and prevents autonomous harmful actions.

Answer

Fully autonomous agent loop with max iterations set to 50

Answer

Azure API Management rate limiting policy

Answer

Asynchronous event grid notification after tool execution

Question 10

Which TWO actions should you take to ensure that your Azure AI deployment is secure?

Accepted Answer

Use Managed Identities for service-to-service communication.. Option B is correct because Managed Identities let Azure AI services authenticate to other Azure resources (such as Key Vault, Storage, or Cognitive Services) through Microsoft Entra ID, eliminating the need to embed credentials or API keys in code and thus reducing the risk of secret leakage. Option D is correct because enforcing the latest TLS version (TLS 1.2 or 1.3) for all communication encrypts data in transit and protects against known protocol downgrade and cipher-suite vulnerabilities. Option A is wrong because leaving a public IP exposed at all times unnecessarily widens the attack surface; access should be restricted via private endpoints, VNet integration, or firewall rules. Option C is wrong because storing keys in a public GitHub repository exposes secrets to anyone, leading to credential compromise. Option E is wrong because assigning the default 'Owner' role to all users violates least privilege and grants excessive control over the Azure AI deployment.

Answer

Keep the public IP exposed at all times.

Answer

Store all keys in a public GitHub repository.

Answer

Use the default 'Owner' role for all users.

Question 11

Which of the following is a core principle of Responsible AI in Azure AI Foundry?

Accepted Answer

Transparency and explainability.. Transparency and explainability is a core principle of Responsible AI in Azure AI Foundry, ensuring that AI systems are understandable, their decisions can be explained, and users can comprehend how outputs are generated. This aligns with Microsoft's Responsible AI principles: fairness, reliability & safety, privacy & security, inclusiveness, transparency, and accountability. The other options describe anti-patterns that violate Responsible AI.

Answer

Bypassing all safety filters.

Answer

Maximizing cost savings at any cost.

Answer

Forcing the model to agree with the user.

Question 12

You are building an advanced agentic RAG solution. When a user asks a complex question, the agent needs to generate multiple search queries, execute them in parallel against Azure AI Search, synthesize the findings, and then formulate a final answer. Which agent pattern are you implementing?

Accepted Answer

Query decomposition and parallel retrieval. Query decomposition and parallel retrieval is an agentic RAG pattern where the agent breaks a complex question into multiple sub-queries, executes them concurrently against Azure AI Search, and then synthesizes the results into a final answer. This pattern improves recall and handles multi-hop reasoning. It is distinct from fine-tuning, keyword extraction, or static templating, which do not involve dynamic query generation and parallel execution.

Answer

Model fine-tuning

Answer

Single-shot keyword extraction

Answer

Static prompt templating

Question 13

Your application integrates Azure AI Vision and processes images stored in an Azure Blob Storage container. To adhere to security best practices, you want to avoid exposing public blob URLs. How should you provide the image to the Image Analysis API?

Accepted Answer

Download the image bytes in your application code and send them in the binary request body (application/octet-stream).. The Image Analysis API accepts image bytes directly in the request body with Content-Type: application/octet-stream, so your application can read the blob using its own credentials (managed identity, SAS, or connection string) and forward the bytes without ever making the blob publicly accessible. This keeps the storage account private and avoids exposing URLs. The API also supports image URLs, but only if the URL is publicly reachable — which violates the stated security requirement.

Answer

Use FTP to transfer images to the vision service endpoint.

Answer

Permanently change the blob container access level to Public Read.

Answer

Mount the Blob Storage container directly to the Azure AI Vision service via NFS.

Question 14

You are developing a solution that uses custom text classification in Azure AI Language. You choose the multi-label classification architecture. What is the key characteristic of multi-label text classification?

Accepted Answer

A single document can be assigned zero, one, or multiple categories simultaneously.. Multi-label text classification in Azure AI Language allows a single document to be assigned zero, one, or multiple categories simultaneously. This is in contrast to multi-class classification, where each document is assigned exactly one mutually exclusive category. The multi-label architecture is ideal when categories are not mutually exclusive and a document can belong to several at once.

Answer

Each document must be assigned to exactly one mutually exclusive category.

Answer

The output is restricted to binary positive or negative values.

Answer

Categories are automatically generated by the service without user training data.

Question 15

You need to ensure that an LLM-based agent in your application follows a specific output schema (JSON) for downstream processing. What is the most reliable way to enforce this?

Accepted Answer

Configure the response_format property to JSON. Configuring the response_format property to JSON (or json_schema for structured outputs) instructs the model at the API level to emit valid JSON conforming to the schema, which is enforced by constrained decoding rather than relying on prompt compliance. This is the most reliable enforcement because it operates at the model's token-generation layer, not on post-hoc parsing.

Answer

Use a few-shot prompt with JSON examples

Answer

Use a custom parser in Prompt Flow

Answer

Set the Stop Sequences parameter