Microsoft Azure AI Fundamentals — Free Practice Questions

Free Microsoft Azure AI Fundamentals practice questions, every answer explained. Covers Azure AI services, computer vision, NLP and responsible AI.

70 questions · answers explained · free to practise · Azure AI certification hub

▶️ Start the interactive quiz

Topics covered

Sample questions with answers

8 of the 70 questions in this set, with the correct answer marked and every option explained.

1. You are tasked to detect faces from a group of individuals. The PersonGroups that you created contain PersistedFace objects of all individuals. You plan to use the recognition models of the Face service in Azure. To compare the accuracy of the models and choose the best model for the intended outcome, you use the parameter recognitionModel in the API call. In the first execution, you choose the default recognition model, and in another execution you use the latest available recognition model. string imageUrl = ""; var faces = await faceClient.Face.DetectWithUrlAsync(imageUrl, true, true, rec Review the scenario above, complete the code for the first model execution by choosing the most appropriate answer choice:

  • recognition_02
    A later recognition model. Switching models invalidates previously enrolled faces, which is precisely the problem when a PersonGroup already holds persisted faces.
  • recognition_03
    Same issue — changing the recognition model requires the PersonGroup to be retrained and its faces re-enrolled.
  • recognition_04
    The newest model in the family, and the same constraint applies: it cannot be mixed with faces enrolled under a different model.
  • recognition_01
    The default recognition model is recognition_01.

2. You create a new LUIS application in the LUIS portal by providing the values for name, language,description and prediction resource. You populate the domain with intents, entities and utterances. Next, you train your application, create a prediction resource and publish the application to an endpoint URL. While you query the endpoint URL for various utterances, you find the top intent and next intent scores are close enough. You also find a few utterances that are not predicted for the labeled intent. Given the scenario above, what are the three options you would use to improve the prediction accuracy?

  • Enable active learning, capture endpoint queries and relabel entities.
    You can review dashboard analysis to find unclear predictions marked with Orange color and incorrect predictions marked with Red color. You can log user queries by enabling active learning, reviewing the utterances with lower prediction scores and review/modify the intents and entities to improve the model accuracy. In the LUIS portal, you can use a pattern to improve prediction accuracy for utterances.
  • Review dashboard colors and find intents with incorrect or unclear predictions.
  • Add example utterances as pattern, train and publish application again.
  • Set endpoint query parameter log=false, capture endpoint queries and relabel entities.
    This is self-defeating: log=false switches endpoint logging off, so there are no captured queries left to review or relabel. Active learning depends on those logs, which is why the correct step enables logging rather than disabling it.

3. You have created a QnA Maker in Azure and created the QnA pairs. To provide the same answer to a question that is different from the original question, you enable alternate phrasing. Review the steps given below and choose two answer choices for the missing steps: Find the question using the Edit page. __________________________________ Add each new phrasing. ___________________________________ Test the knowledge base.

  • Add metadata to filter the answers.
    Metadata tags let you narrow which QnA pairs are eligible for a given request, for instance by product or region. It filters among answers rather than teaching the knowledge base new ways of asking the same question.
  • Retrain the knowledge base.
    Once you find the question, you need to add alternate phrasing in the Question column. After adding alternate phrasing to questions, you need to retrain the knowledge base.
  • Accept questions suggested by active learning.
    A genuinely useful feature, and closely related — active learning proposes alternate phrasings mined from real user queries. But it is a separate, automated workflow; this question asks about the manual steps for adding phrasings you have written yourself.
  • Select add alternate phrasing in the Question column.

4. To help customers locate products and associated insights conveniently, an e-commerce website plans to enable a functionality so that the customers can upload the picture of the product they are looking for and get - Products that are visually similar - Related webpages with similar images - Details of the related entities Which Azure Cognitive Service should be preferred?

  • Video Search
    Returns video results for a text query. It does not take an image as the query input.
  • Entity Search
    Returns information about people, places and things from a text query — not a visually similar product.
  • Image Search
    Finds images matching a text query. Visual Search is the one that accepts an image and finds visually similar items, which is what an image-driven product lookup needs.
  • Visual Search
    visual search lets us upload an image or use a URL to perform a search and get information about it as required in the scenario.

5. Which of the cognitive Service API would help in: - Scanning text to identify personal data - Using a custom list in-line with content policies to block or allow content.

  • Text Protector
    Not an Azure Cognitive Service. The name is plausible but no such API exists.
  • Content Moderator
    Moderation API is included in the content moderator service and helps check materials content for potential inappropriateness or objectionability.
  • Computer Vision
    Analyses images and video — object detection, OCR, description. It does not moderate text against custom term lists.
  • Text Analytics
    Extracts sentiment, key phrases, language and named entities including some personal data types, but offers no custom blocklist or moderation workflow.

6. You are asked to ensure the adherence to compliance of the ML experiment you are working on and ensure proper storing of the related experiment documentation. Which Azure offering will you plan to use?

  • Compliance Manager
    Compliance Manager is a workflow-based Microsoft tool that helps us ensure compliance adherence.
  • Microsoft Service Trust Portal
    The closest option, and easy to confuse. The Service Trust Portal is where you obtain Microsoft's own audit reports and certifications — evidence about Microsoft's compliance. Compliance Manager is where you assess, track and document your organisation's compliance posture.
  • Azure Sentinel
    A cloud-native SIEM for detecting and investigating security threats. It handles incidents, not regulatory documentation.
  • Security centre
    Now Microsoft Defender for Cloud. It assesses security posture and recommends hardening, which overlaps with compliance reporting but is oriented to security configuration rather than experiment documentation.

7. Which Azure cognitive service uses OCR to detect text in image?

  • Computer Vision
    Computer vision is an Azure Cognitive Service that uses OCR to detect text in images.
  • Video indexer Option
    Extracts insights from video, and does include OCR on individual frames. The question concerns a still image, which is Computer Vision's territory.
  • Form Recognizer
    Now called Document Intelligence. It uses OCR but is specialised for pulling structured key-value pairs and tables out of forms and invoices, rather than general text detection in an arbitrary image.
  • Text Detect
    Not an Azure service. Text detection is a capability within Computer Vision, not a product in its own right.

8. You are working on a project in a financial organization and have to adhere to the laid regulatory requirements. The project you are working for involves 3 applications on a Windows server and is using RDBMS. To be compliant with the regulatory requirements you need to ensure that the user data remains in their specific geographic location and is secured at the level of database rows. Which database offering from Azure would be an ideal fit for this scenario?

  • Azure Data Lake
    Object storage optimised for large-scale analytics. It is not a relational database and offers no row-level security for RDBMS workloads.
  • SQL Server on Azure Virtual Machines
    SQL Server on Azure Virtual Machines facilitates the use of full versions SQL Server in the cloud. Manage of hardware is not required as this is a managed offering with pay as you go pricing model. Azure VMs run in different regions across the globe and can be configured to adhere to the regulatory requirements.
  • Azure SQL Edge
    A small-footprint SQL engine built for IoT edge devices. Wrong fit for three applications running on a Windows server.
  • Cosmos DB
    A globally distributed NoSQL database. It can pin data to selected regions, but the scenario specifies an existing RDBMS and row-level security — a SQL Server feature that Cosmos DB does not provide.

62 more questions in the app

Practise the full 70-question set with a timer, scoring and progress tracking.

Start the free quiz
Get the ad-free PRO app

More practice sets

Browse every quiz, tutorial and interactive AI tool on the All Tutorials & Tools page, or jump to a certification hub: AWS, Azure AI, Google Cloud, AWS Data Engineer.