Microsoft Certified: Azure AI Fundamentals practice questions

Question 1

A team wants to build an AI agent that orchestrates multi-step reasoning, manages memory, and uses tools through frameworks. Which approach does Azure OpenAI support for building orchestrated agents?

Accepted Answer

Agent orchestration using frameworks like Semantic Kernel and LangChain for multi-step reasoning, tool use, and memory management

Answer

Content filtering with custom severity configurations

Answer

Batch processing for asynchronous large-scale inference

Answer

Structured outputs through JSON mode for reliable formatting

Question 2

A platform team is migrating a product catalog and FAQ content into Azure AI Search so users can run AI-powered queries over it. The source data lives in a SQL database with mixed column types. What must the team do to make this content searchable?

Accepted Answer

Serialize the data into JSON, then push the documents or pull them with an indexer

Answer

Upload the raw SQL tables directly because Azure AI Search indexes relational rows natively

Answer

Convert the content into PDF files because the index only accepts binary document formats

Answer

Store the data as plain CSV because Azure AI Search rejects any structured serialization

Question 3

An MLOps team is rolling out version 2 of a model that already serves live predictions over HTTPS, and wants to divert a small share of incoming requests to the new version, then increase that share once it proves reliable. Which Azure Machine Learning capability supports this gradual rollout?

Accepted Answer

Split live traffic across multiple deployments on one online endpoint, raising the new version's share gradually

Answer

Deploy the new model to a batch endpoint so scoring jobs run asynchronously in parallel on dedicated compute clusters

Answer

Register each model version in the model registry and let Designer automatically promote the active version

Answer

Run an automated ML job whose ensemble iterations replace the current model once training completes

Question 4

A department is condensing long quarterly reports by pulling out the most relevant original sentences, scoring each sentence for relevance so results can be returned in document order or ranked by importance. Which Azure AI Language summarization approach provides this?

Accepted Answer

Extractive summarization, which ranks original sentences and returns them by position or by rank

Answer

Abstractive summarization, which writes brand-new sentences that paraphrase the original source content

Answer

Conversation summarization, which recaps spoken dialogue into timestamped chapter segments

Answer

Key phrase extraction, which only returns short topic phrases instead of full ranked sentences

Question 5

A developer is enriching an image-accessibility feature and needs one-sentence descriptions of up to 10 distinct regions within each image, each with its own bounding-box coordinates, in addition to one description of the whole image. Which Image Analysis 4.0 feature should the developer enable?

Accepted Answer

Dense Captions, which describe up to 10 regions in one sentence each and return each region's bounding box

Answer

Caption, which generates only a single one-sentence description covering all of the image contents at once

Answer

Tags, which list recognizable objects and actions in the image but never include any bounding-box coordinates

Answer

Object detection, which returns bounding boxes for objects but no natural-language sentence descriptions of regions

Question 6

Which Azure service provides enterprise access to OpenAI models such as GPT-4o, DALL-E 3, and embedding models through Microsoft's cloud infrastructure?

Accepted Answer

Azure OpenAI Service

Answer

Azure Machine Learning

Answer

Azure AI Language

Answer

Azure AI Vision

Question 7

A company is evaluating a credit-scoring model and finds it approves qualified applicants from one demographic group far more often than equally qualified applicants from another group. In Microsoft's responsible AI guidance, this best illustrates which type of AI-caused harm?

Accepted Answer

Harm of allocation, where the system withholds opportunities or resources from certain groups

Answer

Harm of quality-of-service, where the system simply responds more slowly for some users

Answer

A privacy and security harm, where personal training data leaks to unauthorized outside parties

Answer

A transparency harm, where the model cannot produce any explanation for its scored outputs

Question 8

A company needs to process a large backlog of pre-recorded customer service calls for transcription with high throughput. Which Azure Speech processing mode is best suited?

Accepted Answer

Batch transcription for processing large volumes asynchronously

Answer

Real-time transcription for streaming audio

Answer

Fast transcription for individual pre-recorded files

Answer

Pronunciation assessment for evaluating speech quality

Question 9

A customer support platform needs to build a virtual assistant that understands user intentions from natural language input and extracts important information. Which Azure Language capability is designed for this?

Accepted Answer

Conversational Language Understanding for predicting intentions and extracting information from utterances

Answer

Sentiment analysis for determining customer satisfaction

Answer

Key phrase extraction for identifying important terms

Answer

Entity linking for disambiguating entity references

Question 10

A hospital wants to export a trained Custom Vision model to run on medical devices that operate offline in remote clinics. Which export capability does Custom Vision support for edge deployment?

Accepted Answer

ONNX export for edge devices and Docker containers for deployment

Answer

REST API endpoints that work without internet

Answer

Azure Blob Storage packages for offline hosting

Answer

Python serialization files for direct model storage

Question 11

A company needs guaranteed, predictable throughput with flat-rate pricing for their consistently high-volume Azure OpenAI application. Which deployment type should they choose?

Accepted Answer

Provisioned Throughput Managed deployment

Answer

Global Standard deployment

Answer

Batch deployment

Answer

Data Zone deployment

Question 12

An enterprise is building an intelligent document processing pipeline that needs to handle invoices, receipts, and business cards. The solution must extract text from scanned images, identify key entities in the extracted text, and route documents based on content classification. Which combination of Azure AI workload types does this pipeline require?

Accepted Answer

Computer vision OCR for text extraction, natural language processing for entity recognition, and custom text classification for routing

Answer

Speech recognition for audio input, sentiment analysis for prioritization, and generative AI for document summarization

Answer

Image classification for sorting, face detection for identity verification, and translation for multilingual support

Answer

Object detection for layout analysis, speech synthesis for accessibility, and custom vision for brand detection

Question 13

A company has trained multiple model versions and needs to track, compare, and roll back to previous versions if new models underperform. Which Azure ML capability supports this?

Accepted Answer

Azure Machine Learning model registry for versioning

Answer

Azure Machine Learning AutoML for automated selection

Answer

Azure Machine Learning compute clusters for parallel training

Answer

Azure Machine Learning managed endpoints for deployment

Question 14

A news aggregation service needs to link mentions of entities in articles to their correct meanings, distinguishing between Apple Inc. and apple the fruit. Which NLP capability resolves this ambiguity?

Accepted Answer

Entity linking to disambiguate entity identity by returning links to Wikipedia articles

Answer

Named entity recognition to detect the entity mention

Answer

Key phrase extraction to identify the main topic

Answer

Language detection to determine the article's language context

Question 15

A company wants to build a content moderation system that filters user-uploaded images for inappropriate content, allows custom content policies, and supports automated detection with human review for borderline cases. Which Azure approach provides all of these?

Accepted Answer

Azure Content Safety with custom categories and image moderation workflows combining automated detection with human review queues

Answer

Azure AI Vision Image Analysis with tagging and caption generation

Answer

Azure Custom Vision for training custom inappropriate content classifiers

Answer

Azure AI Face service with emotion detection and age estimation

Microsoft Certified: Azure AI Fundamentals

Sample questions

Sources

Similar exams