AI-102 Practice Questions

Projection Groups: In an Azure Cognitive Search knowledge store definition, the projections property is an array where each element represents a projection group. The provided JSON shows the projections property as an array containing a single object (projections: [ { ... } ]). This single object defines all the projections (tables, objects, files) for that group. Therefore, there is only one projection group.

Image Saving: The files projection is specifically used to save files, such as the original or normalized images, into Azure Blob Storage. In the given JSON, the files array is empty ("files": []). This indicates that no configuration exists to save the image files. While the objects projection saves JSON data extracted from the images (specifically, layoutText from normalized_images) to a blob container, it does not save the image files themselves. Thus, the images will not be saved.

Microsoft Documentation: "Create a knowledge store in Azure Cognitive Search":

This document states, "The projections property is an array. You can have more than one projection group." It clarifies that each element in the projections array constitutes a distinct projection group. The sample code shows one element in the array, confirming one projection group.

Under the section for "files projections", it explains, "A files projection... is used when you want to save enriched content, such as image files... to blobs." The absence of any definition within the files: [] array in the question's JSON confirms no image files will be projected (saved).

Microsoft Documentation: "Knowledge store projections":

This page details the different projection types. It defines the files projection as the mechanism to "save the content of a node from an enrichment tree as a file in Azure Storage." Since the files array in the provided code is empty, this action will not occur for any image files.

Custom question answering, a feature of Azure AI Language, is specifically designed to create a conversational question-and-answer layer over existing content, such as product manuals. This service minimizes development effort by allowing you to simply point to your content; it then automatically extracts and indexes information to build a knowledge base. This purpose-built solution is the most direct, cost-effective, and low-effort approach to creating a support chatbot from a single document, directly aligning with the stated requirements.

A. Azure AI Phi-3-medium with fine-tuning requires significant data preparation, training, and deployment effort, which contradicts the goal of minimizing development and cost.

C. This describes a Retrieval-Augmented Generation (RAG) pattern. While powerful, it is more complex and costly to implement, involving multiple services (Azure OpenAI, Azure AI Search), than the purpose-built Custom question answering feature.

D. Azure AI Document Intelligence is used for extracting text and structure from documents. It is a data extraction tool, not a complete chatbot or question-answering solution.

1. Microsoft Documentation, Azure AI Language, "What is question answering?": "Question answering provides cloud-based Natural Language Processing (NLP) that allows you to create a natural conversational layer over your data. It is used to find the most appropriate answer for any input from your custom knowledge base (KB) of information." It lists "Product Manuals" as a common data source.

Source: Microsoft Learn, learn.microsoft.com/en-us/azure/ai-services/language-service/question-answering/overview (Section: "When to use question answering").

2. Microsoft Documentation, Azure AI Language, "Question answering quickstart": The quickstart guide demonstrates that a user can create a fully functional knowledge base simply by providing a URL or uploading a file, highlighting the minimal development effort required.

Source: Microsoft Learn, learn.microsoft.com/en-us/azure/ai-services/language-service/question-answering/quickstart/sdk (Section: "Create a question answering project").

3. Microsoft Documentation, Azure OpenAI, "Azure OpenAI on your data": This document outlines the architecture which requires setting up an Azure AI Search resource, creating an index, and configuring an Azure OpenAI model. This multi-step process is more complex than the single-service approach of Custom question answering.

Source: Microsoft Learn, learn.microsoft.com/en-us/azure/ai-services/openai/concepts/use-your-data (Section: "How it works").

4. Microsoft Documentation, Azure AI Document Intelligence, "What is Azure AI Document Intelligence?": "Azure AI Document Intelligence is a cloud-based Azure AI service that enables you to build intelligent document processing solutions... It's not a standalone conversational agent."

Source: Microsoft Learn, learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview (Section: "What is Azure AI Document Intelligence?").

The first requirement is to convert a customer's spoken words (a call) into text that can be used for a database query. This process is known as speech-to-text or speech recognition. The SpeechRecognizer class in the Azure AI Speech SDK is specifically designed for this purpose, capturing audio input and transcribing it into text.

The second requirement is to provide a spoken response to the customer based on the retrieved order details. This process is called text-to-speech or speech synthesis. The SpeechSynthesizer class is the correct tool for this task, as it converts a text string into audible, synthesized speech.

The other options are incorrect for this scenario: TranslationRecognizer is used for speech translation (e.g., spoken Spanish to written English), and VoiceProfileClient is used for Speaker Recognition to identify who is speaking, not what they are saying.

Microsoft. (2023). Speech-to-text overview. Azure AI services documentation. Retrieved from Microsoft Docs.

Reference Details: This document explicitly states, "Speech-to-text...transcribes audio streams to text." It further identifies the Speech SDK and the SpeechRecognizer object as primary tools for this function.

Microsoft. (2024). Quickstart: Recognize speech from a microphone. Azure AI services documentation.

Reference Details: In the C#, Python, and Java examples, the core object instantiated for speech recognition is new SpeechRecognizer(speechConfig). This confirms its role in converting speech to text.

Microsoft. (2023). Text-to-speech overview. Azure AI services documentation.

Reference Details: This document defines the service: "Text-to-speech...enables your applications, tools, or devices to convert text into humanlike synthesized speech."

Microsoft. (2024). Quickstart: Synthesize speech to a speaker. Azure AI services documentation.

Reference Details: The code samples provided in this guide directly use the SpeechSynthesizer class (e.g., new SpeechSynthesizer(speechConfig)) to convert text strings into spoken audio output.

The question specifies that a query key (read-only) is compromised and explicitly states the requirement to "ensure that users only have read only access." Admin keys (Primary/Secondary) provide full read-write access and should not be used for client apps, as this violates the principle of least privilege.

To rotate a query key with zero downtime, you must follow this specific lifecycle:

1. Add a new query key: unlike Admin keys which are "regenerated," query keys are created as new items in a list (up to 50 per service).
2. Change the app to use the new key: Update the application configuration to direct traffic to the valid key.
3. Delete the compromised key: Once traffic has migrated, remove the old key to revoke unauthorized access.

Microsoft Learn - Connect to Azure AI Search using API keys:

"Query keys are used for read-only access to documents within an index... You can create up to 50 query keys per service." "The process for regenerating a query key is to delete and then recreate it."

Source: Manage Azure AI Search API Keys

Microsoft Learn - Azure CLI Command Reference (Query Keys):

"Unlike admin keys, query keys are not regenerated. The process for regenerating a query key is to delete and then recreate it."

Source: az search service query-key delete

Microsoft Learn - Security Best Practices:

"If you want to continue using API keys, be sure to always monitor who has access to your API keys and regenerate API keys on a regular cadence." (Context: Rotation involves overlap to prevent downtime).

Source: Security controls for Azure AI Search

To add or update synonyms using the Custom Question Answering Authoring API, you must construct a JSON payload that conforms to the service's REST API schema. According to the official documentation, the Update Synonyms operation requires a request body where the top-level property is named "value". This property contains an array of synonym set objects. Each object within this array defines a primary word and its alternatives. The key for the primary word is "value" (e.g., "testing"), and the key for the array containing the list of its synonyms is "alterations" (e.g., ["Trials", "Evaluate"]).

Microsoft Azure AI Language REST API Documentation: The official documentation for the Projects - Update Synonyms operation specifies the exact JSON structure for the request body.

Source: Microsoft Learn, Azure AI Language REST API Reference (API version 2023-04-01).

https://learn.microsoft.com/en-us/rest/api/language/2023-04-01/projects/update-synonyms (Refer to the 'Request Body' section and the 'Alterations' schema definition).

The Computer Vision analyze images service is the appropriate endpoint as it offers a comprehensive suite of pre-trained models for various image analysis tasks, including image type detection. The goal is to classify images into broad types like photographs, drawings, or clipart, which is a standard feature of this service and does not require a custom-trained model.

The imageType property within the API's JSON response is the most precise and direct feature for this task. This property specifically contains indicators (clipArtType and lineDrawingType) that return confidence scores to determine if an image is clipart or a line drawing. Images that do not register as either can be categorized as photographs.

Microsoft Azure Documentation, "Call the Image Analysis API": This official documentation details the capabilities of the Analyze Image operation. Under the "Image type detection" section, it explicitly states, "The Analyze Image operation can detect the content type of images, indicating whether an image is clip art or a line drawing."

Source: Microsoft Learn, Azure Cognitive Services for Vision documentation.

Section: "Image type detection" under the "Call the Image Analysis API" article.

Microsoft Azure Documentation, "ImageAnalysisResult Class": The API reference documentation for the Vision SDK confirms the structure of the analysis result. It shows that the ImageAnalysisResult object contains the imageType property, which in turn holds the clipArtType and lineDrawingType confidence values.

Source: Microsoft Learn, .NET SDK reference for Azure.AI.Vision.ImageAnalysis.

Specifics: ImageAnalysisResult.ImageType property documentation.

The document processing requirement states that a manual review is necessary if a key field's confidence score is less than 70 percent (0.7). The provided JSON output from the Azure AI Document Intelligence (formerly Form Recognizer) API shows the extracted MerchantName field has a confidence score of 0.65. The expression documentResults.fields.MerchantName.confidence < 0.7 correctly evaluates this condition, as 0.65 is less than 0.7. This expression accurately targets the specific field with low confidence, thereby triggering the required manual review process as per the requirements.

A. This expression only verifies that the document is a receipt; it does not evaluate any confidence scores to trigger a manual review.

B. This expression is syntactically invalid. It does not correctly reference the confidence property of any specific field within the fields object.

C. This expression uses the incorrect comparison operator (>). It checks for high confidence, which is the opposite of the requirement to flag low-confidence results for review.

1. Microsoft Azure AI Document Intelligence Documentation, "Prebuilt receipt model": This document outlines the JSON structure returned by the prebuilt receipt model. It confirms that each extracted field, such as MerchantName, has an associated confidence property, which is a numeric value between 0 and 1. The structure documentResults.fields.MerchantName.confidence is the correct path to access this value.

Source: Microsoft Learn, Azure AI services documentation.

Reference: https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-receipt?view=doc-intel-4.0.0, see the "Data extraction" section for field schema.

2. Microsoft Azure AI Document Intelligence Documentation, "Model quality and confidence scores": This official documentation explains the purpose of confidence scores and recommends their use for validation. It states, "The confidence score indicates the probability that the extracted value is correct... For scenarios that are sensitive to accuracy, we recommend that you use a human-in-the-loop review." This directly supports the question's premise of using a confidence threshold (70%) to trigger a manual review.

Source: Microsoft Learn, Azure AI services documentation.

Reference: https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/concept-model-overview?view=doc-intel-4.0.0#model-quality-and-confidence-scores

3. Microsoft Azure AI Document Intelligence SDK Quickstart: The official SDK quickstarts provide code examples demonstrating how to access the analysis results. The code consistently shows accessing a specific field's confidence score by navigating through the result object to the specific field name, such as document.Fields["MerchantName"].Confidence (in C#), which corresponds to the JSON path in the question.

Source: Microsoft Learn, Azure AI services documentation.

Reference: https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/quickstarts/get-started-sdks-rest-api?view=doc-intel-4.0.0&pivots=programming-language-csharp, see the section "Analyze a document from a URI".

The question refers to provisioning a "QnA Maker service" and explicitly mentions an "App Service plan," which points to the classic QnA Maker architecture (prior to its integration into the Azure AI Language service). In this architecture, creating a QnA Maker service automatically provisions several dependent Azure resources.

1. Azure Cognitive Search (D): This service is mandatory. It is used to index the knowledge base content (question-and-answer pairs) and provides powerful search and ranking capabilities to find the most relevant answer for a user's query.

2. Azure App Service (E): This resource is created to host the prediction endpoint. When a client application sends a query, it is this App Service that processes the request, queries the Cognitive Search index, and returns the answer. The pre-existing App Service Plan (AP1) provides the underlying compute resources for this App Service.

A. Language Understanding: This is a separate Azure AI service for intent recognition and entity extraction. It is not a required dependency and is not automatically created with QnA Maker.

B. Azure SQL Database: QnA Maker does not use Azure SQL Database for storing its knowledge base. It relies on Azure Cognitive Search for data persistence and retrieval.

C. Azure Storage: While Azure Cognitive Search may use Azure Storage internally, an Azure Storage account is not one of the primary, user-facing resources that is directly and automatically provisioned as a dependency for the QnA Maker service itself.

1. Microsoft Learn, "What is question answering?": Under the section "Service architecture," the documentation explains that a question answering solution consists of an Azure Cognitive Search resource for storing and indexing data. While this document describes the new architecture (as a feature of Azure AI Language), the fundamental dependency on Cognitive Search remains from the classic QnA Maker service. The question's mention of an App Service Plan points specifically to the classic architecture where an App Service was also a key component.

2. Microsoft Learn, "Migrate from QnA Maker to question answering": This document explicitly discusses the resources in the classic QnA Maker service, stating, "Previously, the QnA Maker service held both the question answering configuration and the query endpoint." This endpoint was hosted on an Azure App Service, which ran on an App Service Plan. This confirms the components of the architecture implied by the question.

To interact with the Azure AI Content Safety service using the .NET SDK, you must first create an instance of the ContentSafetyClient. This class serves as the primary entry point for making calls to the service, and its constructor requires the service endpoint Uri and an AzureKeyCredential.

Next, to prepare the text for analysis, you create an instance of AnalyzeTextOptions. The constructor for this class takes the string of text that needs to be evaluated. This object is then passed to the AnalyzeText method of the ContentSafetyClient instance to perform the content moderation analysis.

Microsoft Azure AI Services Documentation, "Quickstart: Use the Azure AI Content Safety client library for text analysis": This official quickstart guide demonstrates the exact pattern shown in the question. It explicitly shows creating a ContentSafetyClient and then passing an AnalyzeTextOptions object to the AnalyzeText method.

Reference: See the "Authenticate the client" and "Analyze text for harmful content" sections in the C# examples.

Microsoft, .NET SDK Reference, Azure.AI.ContentSafety.ContentSafetyClient Class: The official class documentation confirms that ContentSafetyClient is the main client for accessing the Content Safety service.

Reference: Azure.AI.ContentSafety Namespace documentation.

Microsoft, .NET SDK Reference, Azure.AI.ContentSafety.AnalyzeTextOptions Class: The class documentation specifies that AnalyzeTextOptions is used to provide the text and options for an analysis request.

Reference: Azure.AI.ContentSafety Namespace documentation, constructor details.

To use your own files as grounding data for an Azure OpenAI model, you implement the Retrieval-Augmented Generation (RAG) pattern using the "Azure OpenAI on your data" feature. This feature requires two primary components: a data source to store the files and a search service to index and retrieve information from them. Azure Blob Storage serves as the repository for your unstructured data files (e.g., PDFs, DOCX). Azure AI Search is then used to create a searchable index of the content within those files, enabling the system to find relevant information to ground the model's responses.

A. Azure AI Bot Service: This service is used to build and deploy conversational bots. It can consume the grounded OpenAI model but is not required to create the grounding data source itself.

B. Azure SQL: This is a relational database service for structured data. While it can be a data source, it is not the appropriate service for uploading and storing unstructured files as specified.

C. Azure AI Document Intelligence: This service extracts text and structure from documents. It can be used in a pre-processing step but is not a mandatory resource for the core "on your data" feature.

1. Microsoft Learn, "Azure OpenAI on your data." This document explicitly states the required resources. Under the "How it works" section, it details the data ingestion process: "Azure OpenAI on your data works by ingesting data from a specified data source (such as Azure Blob Storage) and creating a search index in Azure AI Search."

Source: Microsoft. (2024). Azure OpenAI on your data. Microsoft Learn. Retrieved from https://learn.microsoft.com/en-us/azure/ai-services/openai/on-your-data-overview (Refer to the "How it works" and "Data sources" sections).

2. Microsoft Learn, "Quickstart: Chat with Azure OpenAI models using your own data." The prerequisites for the tutorial clearly list the necessary components.

Source: Microsoft. (2024). Quickstart: Chat with Azure OpenAI models using your own data. Microsoft Learn. Retrieved from https://learn.microsoft.com/en-us/azure/ai-services/openai/use-your-data-quickstart?tabs=command-line&pivots=programming-language-python (Refer to the "Prerequisites" section, which lists an Azure AI Search resource and a storage account with Blob Storage).

3. Microsoft Learn, "Data source options for Azure OpenAI on your data." This page provides a table of supported data sources.

Source: Microsoft. (2024). Data source options for Azure OpenAI on your data. Microsoft Learn. Retrieved from https://learn.microsoft.com/en-us/azure/ai-services/openai/concepts/use-your-data#data-source-options (The table explicitly lists "Azure Blob Storage" as a source and notes that "Azure AI Search" is required for indexing).