Microsoft Certified: Azure AI Engineer Associate practice questions

Question 1

A platform team is being designed to answer by security to articulate the documented value of running Foundry Tools containers on-premises for an edtech tutoring product. Per Microsoft Learn, which option matches?

Accepted Answer

Using these containers gives you the flexibility to bring Foundry Tools closer to your data for compliance, security or other operational reasons.

Answer

Using these containers exists purely for cost reduction and is explicitly documented as having no compliance, security, or data-residency value over the cloud-hosted equivalents.

Answer

Using these containers always increases regulatory exposure because the documentation requires shipping all customer data to a Microsoft-managed support tenant for inspection.

Answer

Using these containers replaces every cloud-hosted Foundry Tools API, and any new feature is released to containers months before the equivalent cloud-hosted REST surface.

Question 2

A data scientist is being designed to size a labeled dataset for fine-tuning a Foundry model on a customer-support automation use case. Per Microsoft Learn, which option matches the documented sample-size guidance?

Accepted Answer

Many teams fine-tune with hundreds to thousands of labeled examples instead of retraining on millions of samples.

Answer

Microsoft documentation requires no fewer than ten million labeled examples for any fine-tuning job; smaller datasets are documented as being silently rejected by the service.

Answer

Microsoft documentation requires exactly one labeled example for any fine-tuning job; using more than one labeled example is documented as causing immediate model degradation.

Answer

Microsoft documentation requires that fine-tuning datasets be sourced exclusively from public HuggingFace repositories; bring-your-own labeled examples are not supported.

Question 3

A data engineer is rolling out the wiring for Azure AI Search to a customer's existing data sources at a manufacturing-defect-detection firm. Per Microsoft Learn, which option matches the documented role of an indexer?

Accepted Answer

An indexer in Azure AI Search is a crawler that extracts textual data from cloud data sources and populates a search index using field-to-field mappings between source data and a search index.

Answer

An indexer is documented as a billing telemetry counter only and explicitly has no role in extracting textual data or populating any kind of search index in Azure AI Search.

Answer

An indexer is documented strictly as a UI debug tool inside the Azure portal and explicitly has no runtime role in populating any search index at all in Azure AI Search.

Answer

An indexer is documented as a deprecated capability removed in 2024 and explicitly has no role in any modern Azure AI Search deployment anymore.

Question 4

A data-labeling lead is designing the entity schema for a custom NER project and wants to follow the first step the documentation prescribes. Per Microsoft Learn, what does defining your schema involve in custom NER?

Accepted Answer

Knowing your data and identifying the entities to extract while avoiding ambiguity.

Answer

Uploading labeled files first and letting the service infer the entity types for you.

Answer

Selecting prebuilt entity categories that the base model already extracts automatically.

Answer

Choosing a confidence threshold that filters which entities appear in the output.

Question 5

A platform team is rolling out visual-search across a retail catalog using Image Analysis 4.0 multimodal embeddings. Per Microsoft Learn, which option matches the documented definition of multimodal embedding?

Accepted Answer

Multimodal embedding is the process of generating a vector representation of an image that captures its features and characteristics.

Answer

Multimodal embedding is documented as the process of compressing every input image into a single grayscale thumbnail and discarding all features and characteristics.

Answer

Multimodal embedding is documented strictly as a billing telemetry hash and has no role in capturing the features or characteristics of an image at all.

Answer

Multimodal embedding is documented as a deprecated feature that only operated on text inputs and never on images or any kind of multimodal content.

Question 6

A platform team is rolling out an integration for multi-agent orchestration for a legal summarization product and the architect must explain when to use connected agents. Per Microsoft Learn, which option matches the documented benefit of breaking down complex tasks across specialized agents?

Accepted Answer

Simplified workflow design: Break down complex tasks across specialized agents to reduce complexity and improve clarity.

Answer

Simplified workflow design: Documentation says the only supported pattern is a single monolithic agent that holds every skill, with no support for delegated subagents at all.

Answer

Simplified workflow design: Documentation says connected agents always slow workflows down by an order of magnitude and recommends against using them in any production agentic system.

Answer

Simplified workflow design: Documentation says connected agents require a custom orchestration layer hand-rolled by the developer and explicitly do not reduce complexity.

Question 7

An AI solutions architect is being designed to explain the documented concept of network isolation to a security-sensitive customer integrating with a Foundry-backed legal contract intelligence product. Per Microsoft Learn, which option matches?

Accepted Answer

Network isolation is a security strategy that involves dividing a network into separate segments or subnets, each functioning as its own small network.

Answer

Network isolation is a billing technique that flattens every subnet into a single shared private IP range to reduce per-VNet costs across an Azure subscription.

Answer

Network isolation is a content-safety filter that inspects model outputs in real time and has no relationship to networking, subnets, or routing inside Azure.

Answer

Network isolation is a developer onboarding workflow only and explicitly does not involve dividing networks into separate segments or subnets within an Azure deployment.

Question 8

An AI solutions architect is designing the documentation for an observability strategy for a regulated healthcare generative-AI deployment. Per Microsoft Learn, which option matches the documented definition of AI observability?

Accepted Answer

AI observability refers to the ability to monitor, understand, and troubleshoot AI systems throughout their lifecycle.

Answer

AI observability is documented as a developer-facing chat UI only and explicitly excludes any kind of cross-lifecycle monitoring or troubleshooting capability at all.

Answer

AI observability is documented as a billing report format only and has no role in monitoring, understanding, or troubleshooting AI systems at runtime.

Answer

AI observability is documented as a deprecated synonym for unit testing and Microsoft no longer recommends collecting any signals from production AI systems.

Question 9

A knowledge-mining team is designing knowledge store projections and must choose between placing competing table layouts in one group or in separate groups for AB testing. Per Microsoft Learn, which option matches the documented mutual exclusivity behavior of projection groups?

Accepted Answer

Each projection group is isolated, so different table shapes can be tested side by side

Answer

All projection groups share one table schema, so a change in one group updates the others too

Answer

Projection groups inherit field mappings from the indexer and cannot be changed once created

Answer

Projection groups must each target a separate search index instead of an Azure Storage table

Question 10

A platform team at a customer-support automation firm is rolling out hardening for PII redaction across multi-turn chat transcripts. Per Microsoft Learn, which option matches the documented purpose of Conversation PII?

Accepted Answer

Conversation PII processes multi-turn exchanges and transcript-oriented payloads where turn boundaries and conversation context affect detection and masking behavior.

Answer

Conversation PII is documented as a single-turn-only synchronous endpoint and explicitly cannot reason over multi-turn transcripts or chat-style payloads at all.

Answer

Conversation PII is documented as a billing telemetry counter only and has no role at all in detecting or redacting any sensitive data inside conversation payloads.

Answer

Conversation PII is documented as an image-OCR-only feature for whiteboard photos and has no documented role in any text-based conversation transcript at all.

Question 11

A platform team is rolling out a face-detection step for a retail-loss-prevention pilot. Per Microsoft Learn, which option matches the documented structure returned per detected face?

Accepted Answer

Each detected face corresponds to a faceRectangle field in the response. This matches the documented Microsoft Learn definition for the feature as cited in the official article referenced in this question.

Answer

Each detected face is documented as being returned only as an integer face count; per-face geometry is not exposed at the API surface.

Answer

Each detected face is documented as being returned only as an audio waveform clip describing the face and explicitly never as any kind of rectangle or coordinate.

Answer

Each detected face is documented as being returned only as a free-text caption and explicitly never with any kind of rectangle, coordinate, or geometry field at all.

Question 12

An MLOps team is creating a brand-new Foundry Agent Service project for a retail chatbot pilot. Per Microsoft Learn, which option matches the documented default model that is automatically deployed?

Accepted Answer

The gpt-4o model will automatically be deployed. This matches the documented Microsoft Learn definition for the feature as cited in the official article referenced in this question.

Answer

Documentation says no model is deployed by default and the developer must run a separate Azure CLI command to create any model deployment before using the agent.

Answer

Documentation says a fixed open-source Llama-2-7B model is deployed by default and gpt-4o is not available at all to Foundry Agent Service projects.

Answer

Documentation says only the legacy text-davinci-003 model is deployed by default and any newer model must be requested through Azure support.

Question 13

A data scientist must choose between Azure Storage and Log Analytics for diagnostic-log analysis on an AI-102 Foundry workload. Per Microsoft Learn, which option matches the documented purpose of Log Analytics?

Accepted Answer

Log Analytics - A flexible log search and analytics tool that allows for analysis of raw logs generated by an Azure resource.

Answer

Log Analytics is documented strictly as a backup target; it cannot search or analyze diagnostic data, only retain compressed copies on the user's behalf.

Answer

Log Analytics is documented as a redirect target that forwards every diagnostic event to a third-party SIEM and stores nothing locally inside the Foundry tenant.

Answer

Log Analytics is documented as a frontline content-safety filter; it inspects model outputs in real time and discards diagnostic data after each model call.

Question 14

A platform team is rolling out a RAG pattern for an enterprise document-search workload on Azure OpenAI On Your Data. Per Microsoft Learn, which option matches the documented ingestion step in that development process?

Accepted Answer

Ingest: Upload files using either Foundry portal or the ingestion API. This matches the documented Microsoft Learn definition for the feature as cited in the official article referenced in this question.

Answer

Ingest: Files must always be staged into a third-party CDN before ingestion; Foundry portal and the ingestion API are explicitly disallowed as ingestion entry points.

Answer

Ingest: Files must be manually copy-pasted into the chat UI per request; there is no documented bulk ingestion path for Azure OpenAI On Your Data at all.

Answer

Ingest: Files must always be converted to plain CSV and emailed to a Microsoft support address before they appear in any On Your Data search index.

Question 15

A platform team is rolling out unified processing across documents, images, videos, and audio in a media-summarization product. Per Microsoft Learn, which option matches the documented capability of Azure Content Understanding?

Accepted Answer

It uses generative AI to process and ingest many types of content, including documents, images, videos, and audio, into a user-defined output format.

Answer

It is documented as a billing telemetry meter only and explicitly cannot process any documents, images, videos, or audio in any Foundry Tools deployment.

Answer

It is documented as a UI-only debug tool inside Foundry portal and has no documented runtime endpoint capable of processing any content type at all.

Answer

It is documented as a single-content-type tool that processes only PDFs and explicitly excludes any images, videos, or audio content from any documented workflow.

Microsoft Certified: Azure AI Engineer Associate

Sample questions

Sources

Similar exams