AI-102 Designing and Implementing a Microsoft Azure AI Solution Questions and Answers

The scenario:

You are building an in-vehicle application that uses Azure AI Speech for text-to-speech (TTS).

Requirement: Optimize the quality of synthesized voice output.

You must choose the most appropriate SSML (Speech Synthesis Markup Language) attribute.

Option A – style attribute of the mstts:express-as element

Used to apply expressive styles like " cheerful, " " empathetic, " or " angry. "

Helpful for emotion/tone, but not specifically about audio quality for clarity in motor vehicles.

Option B – level attribute of the emphasis element

Controls how strongly certain words are emphasized.

Improves meaning and focus, but does not optimize overall voice quality.

Option C – pitch attribute of the prosody element ✅

The < prosody > element allows adjustments to pitch, rate, and volume.

Adjusting pitch directly affects the naturalness and clarity of the synthesized speech.

Especially useful in noisy environments such as cars, because modifying pitch can make speech stand out better against background noise.

This directly improves perceived quality of voice output.

Option D – effect attribute of the voice element

Used for special voice effects (like whispering).

Not typically used for general clarity or quality improvements.

Correct Answer: C. the pitch attribute of the prosody element

Microsoft References

SSML support in Azure AI Speech

Prosody element in SSML

The requirement is to process images of forms with Azure Form Recognizer (now part of Azure AI Document Intelligence) while ensuring that the image files are not stored in the cloud.

When sending images directly to the Form Recognizer REST API endpoint, you have two options:

Send the file as binary data in the request body (for local file upload).

Send a URL reference (if the file is stored in Azure Blob Storage or another public location).

Since the requirement explicitly states that the image files must not be stored in the cloud, you cannot send a URL. Instead, you must send the raw binary content of the file directly in the request body.

Option A – raw image binary

Correct. This allows you to directly send the local image file to the API without uploading/storing it in the cloud.

Option B – form URL encoded

Incorrect. Form Recognizer does not accept URL-encoded form data.

Option C – JSON

Incorrect. JSON is only used when providing the API with the URL of a cloud-stored document, not when sending the actual file.

Correct Answer: A. raw image binary

Microsoft References

Form Recognizer REST API – Analyze a document

Quickstart: Extract layout and text – Document Intelligence

" If your document is available locally, pass the file as binary data in the request body. "

Comprehensive Detailed Explanation

You need to analyze confidential documents with the Language service while ensuring those documents never leave your premises. The supported pattern for Azure AI Services (Cognitive Services) is to run the Language service container on-premises and send the app’s requests to that local container endpoint. The container performs inference locally; only metering/billing calls are sent to Azure.

To achieve this:

Provision an on-premises Kubernetes cluster that has internet connectivity.

Azure AI service containers running on-prem require outbound access to Azure for metering and billing unless you’re using the special disconnected model (not implied here). Therefore, the cluster must have internet connectivity so the container can reach the Azure billing endpoint while keeping document content local.

Pull an image from the Microsoft Container Registry (MCR).

Official Azure AI service containers (including Language) are published in MCR, not Docker Hub. You must pull the Language (Text Analytics/Language) container image from MCR.

Run the container and specify an API key and the Endpoint URL of the Cognitive Services resource.

When you start the container, you configure it with the API key and billing (endpoint) URL of your Azure Cognitive Services (multi-service or Language) resource. This enables the container to authenticate and report usage to Azure, while your data stays on-prem.

Why not the other options?

AKS would place workloads in Azure, which isn’t required when the requirement is to keep documents on-prem.

On-prem cluster isolated from the internet would block the container’s required calls to Azure’s billing endpoint.

Pull from Docker Hub is incorrect because Azure AI service containers are published to MCR.

App ID and Client Secret are not the standard parameters for Azure AI service containers; they require API key + billing endpoint.

Microsoft Azure AI Solution References

Use Azure AI services in containers (billing/endpoint and API key requirements, data remains local):

Microsoft Learn – Install and run containers for Azure AI services and Use containers with Azure AI services (metering and billing) .

https://learn.microsoft.com/azure/ai-services/containers/

Language service container images on Microsoft Container Registry (MCR):

Microsoft Learn – Deploy language containers (Text Analytics/Language) .

https://learn.microsoft.com/azure/ai-services/language-service/how-to/use-containers

Networking requirements (internet connectivity for metering):

Microsoft Learn – Configure containers and network access for Azure AI services containers .

https://learn.microsoft.com/azure/ai-services/containers/container-requirements-and-limits

These documents collectively show that: (a) containers are pulled from MCR, (b) they are configured with API key + endpoint for billing, and (c) internet egress is required for metering while payload data remains on-premises.

You are securing an Azure Cognitive Search resource with two specific requirements:

Prevent access to Search1 from the internet

The correct way is to use a private endpoint.

A private endpoint connects the Cognitive Search resource directly into your virtual network (VNet), blocking public internet access.

Configuring an IP firewall can restrict access to specific IP ranges but does not fully prevent internet access.

Azure roles and authentication do not address network-level restrictions.

Limit the access of each app to specific queries

This is done via Azure role-based access control (RBAC).

You can assign different Azure roles or API keys with query rights to control what each app can do.

Key authentication alone only provides coarse-grained access (admin/query keys) but does not allow fine-grained per-app restrictions.

RBAC ensures principle of least privilege is enforced.

Correct Answer:

Prevent access from internet → Create a private endpoint

Limit access to queries → Use Azure roles

Microsoft References

Secure access to Azure Cognitive Search using private endpoints

Role-based access control in Azure Cognitive Search

For the category field in Azure Cognitive Search:

searchable → allows users to search by words in the category field.

facetable → enables drill-down filtering (facets).

retrievable → ensures the field appears in search results.

filterable/sortable/key are not required here based on the scenario.

Correct Answer: A

Comprehensive Detailed Explanation

The scenario is identical to Question 43:

You have a FindContact intent.

Training phrases include city names (London, Seattle, Ukraine).

You must model location properly in Language Understanding.

Why the Proposed Solution is Wrong

The solution says: “Create a new entity for the domain.”

Entities in LUIS/CLU represent specific data types to extract (e.g., location, date, contact name).

Domains in Language Understanding are prebuilt sets of intents and entities for common use cases (like Calendar, Home Automation, Communication).

Creating an “entity for the domain” is conceptually incorrect — entities are not meant to represent domains.

What we really need is to model location as an entity (for example, using the prebuilt GeographyV2 entity), not to create a generic “domain entity.”

Correct Approach

Keep the FindContact intent.

Add a Location entity (prebuilt GeographyV2 or custom “Location”).

Train the model so that London, Seattle, Ukraine are recognized as entity values within the FindContact intent.

Option Analysis

A. Yes → Incorrect. This would not meet the goal.

B. No → Correct. Creating a domain entity is not the right way; instead, a proper location entity should be added.

Correct Answer: B. No

Microsoft References

Intents vs Entities in Language Understanding

Prebuilt entity: GeographyV2

LUIS domains overview

 The POST request starts an asynchronous analysis job for video01.mp4. Yes

 You use the URL in the Operation-Location header to poll and retrieve the results . Yes

 The POST does not return the full transcript/key frames in the response body (only 202 Accepted). No

When you invoke a custom analyzer in Azure AI Content Understanding using the analyzers/{analyzerId} endpoint with API version 2024-12-01-preview , the service runs as a long-running operation. The initial response is 202 Accepted and includes an Operation-Location header pointing to the results URL. You then GET that URL to check status and, once complete, obtain outputs (for a video analyzer, this includes transcription and key frames because those capabilities are part of the analyzer definition). The header Ocp-Apim-Subscription-Key is valid for this API.

Microsoft References

Quickstart (REST): asynchronous analyze call returns 202 with Operation-Location ; poll to get results. Microsoft Learn

What’s new: API version 2024-12-01-preview . Microsoft Learn

Prebuilt video analyzer functions (transcript + key frames). Custom analyzers inherit configured tasks. Microsoft Learn

REST API surface: Get Result and related operations for analyzers. Microsoft Learn

Security header definition ( Ocp-Apim-Subscription-Key ). Microsoft Learn

To build a product support chatbot from a product support manual with minimal development effort and cost, use Custom question answering (CQA) in Azure AI Language. CQA lets you ingest files (including manuals/FAQs), automatically extract Q & A, author answers where needed, and expose a ready-to-call endpoint that plugs straight into Azure Bot Service—no fine-tuning or complex retrieval stack required.

Not A (Phi-3 with fine-tuning): fine-tuning adds cost/complexity and isn’t necessary for a manual-based Q & A bot.

Not C (GPT-4 with grounding via Azure AI Search): powerful but more engineering and cost (indexing, RAG orchestration) than needed.

Not D (Document Intelligence): focuses on OCR/structured data extraction, not conversational Q & A.

References (Microsoft Docs):

Custom question answering overview and scenario (create a conversational layer over your data; supports manuals/FAQs).

CQA supports unstructured documents and quick setup in Language Studio/Azure AI Foundry.

Video Indexer supports multiple Person models per account. Once a model is created, you can use it by providing the model ID of a specific Person model when uploading/indexing or reindexing a video. Training a new face for a video updates the specific custom model that the video was associated with.

Note: Video Indexer supports face detection and celebrity recognition for video content. The celebrity recognition feature covers about one million faces based on commonly requested data source such as IMDB, Wikipedia, and top LinkedIn influencers. Faces that aren ' t recognized by the celebrity recognition feature are detected but left unnamed. Once you label a face with a name, the face and name get added to your account ' s Person model. Video Indexer will then recognize this face in your future videos and past videos.

Comprehensive Detailed Explanation

When an indexer runs in Azure AI Search, it processes each document through a fixed series of stages:

Document cracking

The indexer opens the item (file/row) and extracts text, images, and metadata, creating the root of the enriched document (/document). This is always the first stage.

Field mappings

Next, source fields are mapped directly to target index fields before any transformations . Field mappings occur after document cracking but before skillset execution . Microsoft Learn

Skillset execution

Built-in or custom skills (for example, language and sentiment analysis in your scenario) run and write outputs into the in-memory enriched document. Microsoft Learn

Output field mappings

Because skill outputs live only in memory, you must define output field mappings to route selected nodes of the enriched document into index fields. This step comes after skillset execution.

Push to index

Finally, the indexer hands off raw and enriched content to the search index (the physical data structures) so it becomes searchable—i.e., the content is “pushed to the index.” Microsoft Learn

This order— document cracking → field mappings → skillset execution → output field mappings → push to index —ensures raw content is extracted, direct source-to-index mappings are applied, AI enrichment is performed, enriched outputs are projected to fields, and the completed documents are written into the index.

Microsoft Azure AI References

Indexers in Azure AI Search – Stages of indexing (Document cracking; Field mappings; Skillset execution; Output field mappings, with field mappings occurring before transformations). Microsoft Learn

Skillset concepts (Output of skillset is routed to the index via output field mappings after skillset execution). Microsoft Learn

AI enrichment concepts (Field mappings move raw source content to the index; output field mappings move enriched content; indexing ingests raw and enriched content into the physical index). Microsoft Learn

Annotations & enriched document tree (Enriched document is created during document cracking at /document). Microsoft Learn

Defender for Cloud Apps (A): Focused on monitoring SaaS app usage and cloud security, not image content moderation → ❌

Custom Vision (B): Can classify images, but requires building/training your own model → more development effort → ❌

Azure AI Vision (C): Can analyze images, including adult/racy content detection, but Content Safety is now the preferred service for moderation. Still valid → ✅

Azure AI Content Safety (D): Specifically designed for detecting and blocking harmful/inappropriate content in text and images → ✅

Document Intelligence (E): Extracts text/structure from documents, not relevant here → ❌

Correct Answer:

C. Azure AI Vision and D. Azure AI Content Safety

You have files with the following formats:

File1: WMV (34 mins, 400 MB)

File2: AVI (90 mins, 1,200 MB)

File3: MOV (300 mins, 980 MB)

File4: MP4 (80 mins, 1,800 MB)

You must determine which files can be uploaded to Azure AI Video Indexer.

Key Rules (Video Indexer upload limitations):

Supported formats: MP4, MOV, WMV, AVI, M2TS.

Maximum file length: 4 hours (240 minutes).

Maximum file size: 2 GB (2048 MB) for website uploads.

Check each file:

File1 (WMV, 34 min, 400 MB): Supported format, within time and size → ✅ Allowed.

File2 (AVI, 90 min, 1,200 MB): Supported format, within time and size → ✅ Allowed.

File3 (MOV, 300 min, 980 MB): Supported format, but exceeds 240 min limit → ❌ Not Allowed.

File4 (MP4, 80 min, 1,800 MB): Supported format, within size (under 2 GB), within time → ✅ Allowed.

Correct Answer:

A. File1, File2, and File4 only

Microsoft Reference:

Video Indexer upload limitations

Comprehensive Detailed Explanation

The requirement:

You have an Azure AI Document Intelligence resource (formerly Form Recognizer).

You want the app to detect handwritten content inside PDF files.

Step 1 – Choosing the correct model

prebuilt-document: Extracts structure and text from documents, but not optimized specifically for handwriting.

prebuilt-contract: Designed for contracts, not general handwriting.

prebuilt-read: Extracts text (printed and handwritten) from documents. Supports handwritten content recognition.

Therefore, to analyze handwritten content, the correct choice is prebuilt-read.

Step 2 – Confidence threshold

The style.Confidence value indicates the confidence that the text is handwritten.

A typical threshold used is 0.75 (75%).

This ensures that only results with reasonably high confidence are considered handwritten.

Values like 0.1 would be too low (allowing false positives), while 1.0 might miss valid handwriting because it requires absolute certainty.

Correct Answer:

Model: prebuilt-read

Confidence threshold: 0.75

Upload to: An Azure Storage account

Model type: Custom neural

You have 1,000 scanned hand-written survey responses with no consistent layout. The goal is to extract data while minimizing development effort.

Step 1: Upload location

For Azure AI Document Intelligence (formerly Form Recognizer), training data (images, PDFs) must be placed in a container in an Azure Storage account.

Options like Cosmos DB or Azure Files share are not valid training data sources for Document Intelligence.

Therefore, you upload the survey images to Azure Storage account.

Step 2: Model type

The surveys are handwritten and do not have a consistent layout.

Custom template models work only when the layout is fixed and consistent (e.g., invoices with the same fields in the same place).

Identity document (ID) model is a prebuilt model for passports, driver’s licenses, etc., not for surveys.

Custom neural model is designed to handle unstructured or variable-layout forms, including handwritten content. This best fits the survey scenario.

Thus, the correct configuration is:

Upload to Azure Storage account

Use Custom neural model

Correct Answer:

Upload to: An Azure Storage account

Model type: Custom neural

Microsoft References

Azure AI Document Intelligence – Custom models

Custom neural vs custom template models

Train a custom model with Azure Storage input

The requirement is:

" Measure the public perception of your brand on social media using natural language processing (NLP). "

A. Content Moderator

Designed to filter offensive or inappropriate content in text, images, and videos.

Not suitable for sentiment analysis or measuring perception.

B. Form Recognizer

Extracts structured data from documents (invoices, receipts, forms).

Not used for analyzing social media posts.

C. Computer Vision

Analyzes images and videos for object detection, image classification, OCR, etc.

Not used for text-based sentiment analysis.

D. Language service ✅

Provides natural language processing (NLP) capabilities such as sentiment analysis, opinion mining, key phrase extraction, entity recognition , and language detection .

Perfect fit for analyzing public perception from social media posts .

Comprehensive Detailed Explanation along with All References available from Microsoft Azure AI Solution at the end of the explanation

Typing: In a .chat transcript, you can simulate the bot’s typing indicator with the Typing activity, optionally followed by a [Delay=...] . This matches the intent to pause before sending the next message.

AttachmentLayout=carousel: When sending multiple attachments (the two watch images), set the layout to carousel so the Emulator displays the attachments horizontally as a card carousel.

Adaptive Card attachment: The JSON file cards\watchProfileCard.json represents an Adaptive Card; specify adaptivecard so the Emulator renders the card.

Resulting .chat snippet (essentials):

user =User1

bot=watchbot

user : I want a new watch.

bot: [Typing][Delay= 3000 ]

bot: I can help you with that! Let me see what I can find.

bot: Her e ' s what I found.

bot:

[AttachmentLayout=carousel]

[Attachment=https://contoso.blob.core.windows.net/watch01.jpg]

[Attachment=https://contoso.blob.core.windows.net/watch02.jpg]

user: I like the first one.

bot: Sure, pulling up more information.

bot: [Attachment=cards\watchProfileCard.json adaptivecard]

user: That ' s nice! Thank you.

bot: Sure, you are most welcome!

References

Bot Framework Emulator transcript ( .chat ) activities and typing indicator: Bot Framework Activity schema; Emulator supports typing and delay in transcripts.

https://learn.microsoft.com/azure/bot-service/bot-service-debug-emulator

https://learn.microsoft.com/azure/bot-service/rest-api/bot-framework-rest-connector-api-reference#activity-object

Attachment layout ( carousel / list ) for multiple attachments:

https://learn.microsoft.com/azure/bot-service/nodejs/bot-builder-nodejs-send-rich-cards#carousel-layout

Adaptive Cards in Bot Framework:

https://learn.microsoft.com/adaptive-cards/get-started/bots

https://learn.microsoft.com/azure/bot-service/bot-builder-howto-add-media-attachments#adaptive-cards

The app must:

Identify road signs and extract text from images

This is an OCR (Optical Character Recognition) task.

Azure AI Vision provides OCR capabilities to detect and extract text from images (including road signs).

Correct choice: Azure AI Vision.

Analyze the text to identify well-known locations

Once text is extracted, you need to detect if it refers to well-known places (e.g., cities, landmarks).

This is a Named Entity Recognition (NER) problem.

Azure AI Language service includes entity recognition for identifying locations, people, organizations, etc.

Correct choice: Azure AI Language.

Other options:

Azure AI Document Intelligence → Focused on structured documents (invoices, forms), not road signs.

Azure AI Search → Useful for indexing and search, not text extraction or entity recognition.

Correct Answer Matrix:

Extract the text → Azure AI Vision

Identify well-known locations → Azure AI Language

Microsoft References

Azure AI Vision – OCR text extraction

Azure AI Language – Named Entity Recognition (NER)

The question is about evaluating a Custom Vision classifier after training.

When you train an image classifier in Azure Custom Vision, the service automatically calculates performance metrics to help you evaluate the quality of the model.

Metrics Provided by Custom Vision:

Precision:

The percentage of correct positive predictions out of all positive predictions made.

Formula: Precision = TP / (TP + FP)

Helps determine how reliable positive predictions are.

Recall:

The percentage of actual positives that were correctly predicted.

Formula: Recall = TP / (TP + FN)

Helps determine how many of the true positives were captured.

These two metrics are explicitly shown in the Custom Vision portal and via API.

Option Analysis

A. Recall

Yes, Custom Vision reports recall.

Correct

B. F-score

Not directly reported in the Custom Vision portal.

Although it can be derived from precision and recall, it is not provided as a direct metric.

Incorrect

C. Weighted accuracy

Not reported by Custom Vision.

Incorrect

D. Precision

Yes, Custom Vision reports precision.

Correct

E. Area under the curve (AUC)

Not reported by Custom Vision.

More common in ROC curve analysis, not part of Custom Vision output.

Incorrect

Correct Answer: A. recall, D. precision

Microsoft References

Evaluate the prediction performance of your classifier in Custom Vision

Custom Vision training and evaluation

Azure Metrics Advisor builds on Anomaly Detector but adds root cause analysis , correlation across multiple signals, incident detection, and alerting out-of-the-box.

D. Anomaly Detector → Works for single-signal anomaly detection but lacks correlation and root cause analysis.

C. Azure Machine Learning → Would work but requires custom model development (more dev effort).

B. Form Recognizer → Used for document processing, not sensor telemetry.

Correct Answer: A

When you want to debug a chatbot endpoint remotely that was built using the Microsoft Bot Framework, you need tools that allow you to test and tunnel your bot service to your local development environment.

1. Bot Framework Emulator

This is the official testing and debugging tool for bots built with the Microsoft Bot Framework.

It allows developers to:

Connect to a bot running locally or remotely.

Inspect messages sent and received.

Debug conversations and payloads in detail.

Required for local testing of the bot.

Correct.

2. ngrok

Ngrok is a tunneling tool that exposes your local development server (localhost) to the internet via a secure tunnel.

This is essential when debugging remotely because the Bot Framework Service (Azure Bot Service) requires a publicly accessible HTTPS endpoint for communication.

With ngrok, your local bot endpoint (e.g., http://localhost:3978/api/messages ) is mapped to a secure public endpoint (e.g., https://randomstring.ngrok.io/api/messages ).

Correct.

Incorrect Options

A. Fiddler

A web debugging proxy. Useful for monitoring traffic, but not required for Bot Framework remote debugging.

Incorrect.

B. Bot Framework Composer

A tool for building conversational bots visually. Not required for debugging an existing endpoint.

Incorrect.

D. Bot Framework CLI

Used for bot project management and configuration. Not specifically required for remote debugging endpoints.

Incorrect.

F. nginx

A web server/reverse proxy. Not needed for this scenario.

Incorrect.

Correct Answer: C. Bot Framework Emulator, E. ngrok

Microsoft References

Debug a bot locally using Bot Framework Emulator

Use ngrok with Bot Framework Emulator

In SQL, DML (Data Manipulation Language) statements include SELECT , INSERT , UPDATE , and DELETE . Often, you need to filter which rows are affected or returned. That filtering is done with the WHERE clause.

Let’s review the options:

ALTER

Used in Data Definition Language (DDL), not DML.

Alters the structure of a database object (e.g., add/remove columns, modify constraints).

Not used to filter rows.

ON

Used in JOIN operations to specify the relationship between tables.

Does not specify row-level filtering criteria in general DML.

SET

Used in the UPDATE statement to assign new values to columns.

Does not specify which rows to update.

WHERE

Used in DML statements to specify criteria for rows that must match.

Examples:

SELECT * FROM Customers WHERE Country = ' USA ' ;

DELETE FROM Orders WHERE OrderDate < ' 2020-01-01 ' ;

UPDATE Products SET Price = Price * 1.1 WHERE CategoryID = 5;

Thus, the correct clause is WHERE.

Correct Answer: WHERE

Microsoft References

SQL WHERE clause

SQL UPDATE with WHERE

SQL DELETE with WHERE

First blank: speech_recognition_language

Second blank: add_target_language

speech_translation_config = speechsdk.translation.SpeechTranslationConfig(

subscription=os.environ.get( ' SPEECH_KEY ' ),

region=os.environ.get( ' SPEECH_REGION ' )

)

# Source (recognition) language: English (US)

speech_translation_config.speech_recognition_language = " en-US "

# Target translation language: Italian

speech_translation_config.add_target_language( " it " )

audio_config = speechsdk.audio.AudioConfig(use_default_microphone= True )

recognizer = speechsdk.translation.TranslationRecognizer(

translation_config=speech_translation_config,

audio_config=audio_config

)

Comprehensive Detailed Explanation along with All References available from Microsoft Azure AI Solution at the end of the explanation

To translate speech, configure a SpeechTranslationConfig with (1) the recognition language of the input audio and (2) one or more target languages for translation. In Python, you set the input language with the speech_recognition_language property (e.g., " en-US " ), and you add a target with add_target_language( " < locale > " ) (e.g., " it " for Italian). This exactly matches the SDK guidance and examples for speech translation. Microsoft Learn

speech_recognition_language specifies the locale used by the recognizer to understand the spoken input.

add_target_language( " it " ) adds Italian as a translation output language.

Microsoft References

How to translate speech with the Speech SDK (Python examples show translation_config.speech_recognition_language = from_language and translation_config.add_target_language(...) ). Microsoft Learn

SpeechTranslationConfig (Python) reference — add_target_language(language: str) . Microsoft Lea rn

Recognize speech (property usage for speech_recognition_language ). Microsoft Learn

Language support for Speech service (locales and codes for recognition/translation)

You need to minimize indexing time for 10 GB of scanned docs and images.

Azure Cognitive Search supports parallel indexing to increase throughput by partitioning work.

A text-based indexer or scheduled indexing will not reduce build time.

Field mappings help transform fields but don’t affect indexing speed.

Therefore, the correct way to speed up index building is parallel indexing.

Correct Answer: A

Comprehensive Detailed Explanation

To move a Question Answering project to a different Azure region, you migrate the project data and then re-deploy it in the target region:

Export the existing project (source region).

In Language Studio (Question Answering), you first export the project. This produces a package (JSON/ZIP) containing your knowledge bases, sources, synonyms, and settings. Exporting is the supported way to move a project between resources/regions.

Import the project into the new Language service instance (target region).

In the target region’s Language service instance, use Language Studio to import the previously exported file. This recreates the project (KBs and configuration) under the new resource.

Train and publish from the new instance.

After import, you must train (build) and then publish the project so that an endpoint is created under the new regional resource. Clients can then point to the new endpoint and key.

Actions like “enable custom text classification” or “regenerate the keys” are not required for moving a Question Answering project, and training/publishing from the original instance won’t deploy to the target region.

Microsoft Azure AI Solution References

Export and import Question Answering projects in Language Studio (move between resources/regions):

Microsoft Learn – Export and import projects (Question Answering) .

https://learn.microsoft.com/azure/ai-services/language-service/question-answering/how-to/export-import-project

Train and publish a Question Answering project:

Microsoft Learn – Create, train, and publish projects (Question Answering) .

https://learn.microsoft.com/azure/ai-services/language-service/question-answering/how-to/create-test-deploy

Language service overview and regional deployment concepts:

Microsoft Learn – What is Azure AI Language?

https://learn.microsoft.com/azure/ai-services/language-service/overview

 Provide access to AI1 by using: A bearer token

 Connect to the deployment by using: A deployment name

To ensure only your apps can access the Azure OpenAI resource, authenticate with Microsoft Entra ID (Azure AD) and obtain a Bearer token (ideally via managed identity ). This avoids sharing long-lived API keys and lets you scope access by app identity and Azure RBAC. Microsoft’s guidance recommends token-based auth for Azure OpenAI and shows how to acquire a bearer token/managed identity token for calls

When calling the REST API , Azure OpenAI requires you to specify the deployment name to select which model deployment to use. The REST route includes {deployment-id} (deployment name) in the path, e.g., .../openai/deployments/{deployment-id}/chat/completions. This is explicitly documented as a key difference from OpenAI’s public API, which takes only a model.

Therefore:

Access method: Bearer token (via Entra ID/managed identity)

Deployment selector: Deployment name in the REST path

Microsoft References

Authenticate to Azure OpenAI with Microsoft Entra ID / Managed identity (bearer token): guidance and how-to

Azure OpenAI request header recommends token-based authentication (Bearer).

Azure OpenAI requires deployment name in API calls; REST path uses

The question requires enabling the bot to respond to spoken requests with minimal effort. Among the options:

Cortana integration automatically adds speech input/output capabilities to the bot.

Direct Line Speech is another option but requires additional setup and coding, so not the minimal effort solution.

Microsoft Teams is primarily for text-based interactions (though can do voice calls, but not direct bot speech input).

Azure function + Speech would require more development overhead.

Thus, Cortana integration provides the simplest path.

The None intent is a built-in intent in CLU/Language Understanding models.

It is used to capture queries that don’t match any of the defined intents.

By adding representative irrelevant utterances to the None intent , the model learns how to reject spurious queries.

Correct Answer: C

In the Bot Framework SDK, if you want to collect structured input step by step, you typically use a Waterfall dialog. Each step in the waterfall represents one piece of logic, such as asking for the size, toppings, or type of food item.

Prompts are used inside waterfall steps to actually ask the user for input (e.g., text prompt, choice prompt, date-time prompt).

Adaptive, Action, Input are concepts more relevant to Power Virtual Agents or Adaptive Dialogs (Composer), not the Bot Framework SDK question context.

Since the question specifically mentions Bot Framework SDK, the correct answer is Waterfall dialog + Prompts.

Correct Answer: C, D

To access multiple Azure AI capabilities (e.g., Speech and Language) using a single endpoint and a single set of credentials, you should provision an Azure AI multi-service resource (called Azure AI service in the portal; formerly “Cognitive Services” multi-service account). This resource provides one endpoint and one key (or Microsoft Entra ID) that authorizes calls across supported services, including Speech and Language. Creating separate “Azure AI Speech” or “Azure AI Language” resources would give you distinct endpoints/keys for each, which does not meet the requirement. Azure AI Content Safety is unrelated.

Microsoft References

Azure AI services multi-service resource (single endpoint/credential across services).

https://learn.microsoft.com/azure/ai-services/multi-service-resource

What are Azure AI services (overview of consolidated access model).

https://learn.microsoft.com/azure/ai-services/what-are-ai-services

In Power BI, reports are interactive, visual, and support KPIs, drill-through, and drill-down functionality.

Power Apps is for app development, not analytics.

A view is a database object, not a visualization solution.

Dataflow is for ETL/ELT (data preparation), not reporting.

Correct Answer: A

To detect hateful content with Azure AI Foundry Content Safety – Text , you should read the category analysis results returned by the analyze_text call. The response exposes a collection named categories_analysis, where each item contains the analyzed category (e.g., TextCategory.HATE) and its severity level.

Hence:

Iterate over response.categories_analysis, filter the item where item.category == TextCategory.HATE.

Print the severity using hate_result.severity.

This uses category analysis (not blocklist matches), which is the correct mechanism for detecting hate content with a severity score.

Microsoft References

Azure AI Content Safety – Analyze text (Python quickstart/samples): iterate response.categories_analysis, check TextCategory.HATE, read item.severity.

Azure AI Content Safety – Text categories and severity levels: explains categories (Hate, Violence, Sexual, Self-harm) and severity output.

Comprehensive Detailed Explanation

You’re creating a resource that will handle sentiment analysis (Language) and OCR (Vision) with a single key and endpoint. That is a Cognitive Services multi-service (Azure AI services) account.

When creating any Azure resource via ARM (HTTP/REST), the URL must be scoped to a subscription, not a tenant, and use the correct resource provider.

Scope segment: Use

https://management.azure.com/subscriptions/{subscriptionId}/resourceGroups/{rg}/providers/...

Therefore, choose subscriptions/8d3591aa-96b8-4737-ad09-00f9b1ed35ad (not the tenant scope).

Resource provider: Multi-service cognitive account is under Microsoft.CognitiveServices.

The full path to create would be like:

/providers/Microsoft.CognitiveServices/accounts/{accountName}?api-version=2021-10-01

This creates a single Cognitive Services (Azure AI services) resource that provides one endpoint and one key across multiple services (e.g., Language for sentiment analysis and Vision for OCR).

Microsoft Azure AI Solution References

Azure Resource Manager: Create Cognitive Services account (Microsoft.CognitiveServices/accounts) using REST and API version 2021-10-01.

Azure AI services (Cognitive Services) multi-service resource overview—single key/endpoint across services.

The requirement is:

" Measure the public perception of your brand on social media using natural language processing (NLP). "

A. Content Moderator

Designed to filter offensive or inappropriate content in text, images, and videos.

Not suitable for sentiment analysis or measuring perception.

B. Form Recognizer

Extracts structured data from documents (invoices, receipts, forms).

Not used for analyzing social media posts.

C. Computer Vision

Analyzes images and videos for object detection, image classification, OCR, etc.

Not used for text-based sentiment analysis.

D. Language service ✅

Provides natural language processing (NLP) capabilities such as sentiment analysis, opinion mining, key phrase extraction, entity recognition , and language detection .

Perfect fit for analyzing public perception from social media posts .

This question is about selecting valid training files for a custom Form Recognizer model in Azure AI Document Intelligence (formerly Form Recognizer).

Supported File Formats for Training

According to Microsoft documentation, Form Recognizer supports the following for training and analysis:

PDF (≤ 500 pages, ≤ 50 MB for training)

JPEG/JPG (≤ 50 MB)

PNG (≤ 50 MB)

TIFF (≤ 50 MB)

Unsupported:

MP4 → Video not supported.

GIF → Not supported.

Maximum File Size Rules (for training input files):

Images (JPG, PNG, TIFF): ≤ 50 MB

PDFs: ≤ 50 MB

Note: 500 pages limit applies to PDFs.

File-by-File Analysis

File1 – PDF – 20 MB

Type supported.

Size within limit ( < 50 MB).

Valid ✅

File2 – MP4 – 100 MB

MP4 is not supported.

Invalid ❌

File3 – JPG – 20 MB

Type supported.

Size within limit ( < 50 MB).

Valid ✅

File4 – PDF – 100 MB

Type supported.

Size exceeds the 50 MB maximum limit.

Invalid ❌

File5 – GIF – 1 MB

GIF format not supported.

Invalid ❌

File6 – JPG – 40 MB

Type supported.

Size within limit ( < 50 MB).

Valid ✅

Correct Answer:

A. File1, C. File3, F. File6

Microsoft References

Azure AI Document Intelligence (Form Recognizer) – Supported file formats

Train a custom model with Azure Form Recognizer

In Azure Stream Analytics, queries are defined using a SQL-like query language specifically designed for real-time analytics over streaming data.

The language is often referred to as Stream Analytics Query Language (SAQL).

It is very similar to T-SQL (Transact-SQL) but extended with additional features for temporal (time-based) operations, stream joins, aggregations over windows, and event handling.

With this language, you can perform operations like filtering, projections, aggregations, joins, and even machine learning model scoring on data streams.

Reviewing the options:

SQL – Correct. Azure Stream Analytics jobs are defined using a SQL-like language.

XML – Incorrect. XML is used for data representation, not defining queries in Stream Analytics.

YAML – Incorrect. YAML is commonly used for configuration files, not streaming queries.

KOL – Incorrect. This is not a relevant query language.

Correct Answer: A. SQL

Microsoft References

Introduction to Azure Stream Analytics query language

Query overview for Stream Analytics

Comprehensive Detailed Explanation

The method shown uses the Read API of the Azure Computer Vision client library. The ReadAsync call submits the OCR job and returns an Operation-Location header, which contains the operationId. Then GetReadResultAsync(Guid.Parse(operationId)) is called to fetch the result.

The Issue

The Read API is asynchronous.

Submitting the read request does not mean the OCR has finished.

If you immediately call GetReadResultAsync , the operation may still be in progress, resulting in incomplete or empty results.

Required Fixes

Check the results.Status property

The response from GetReadResultAsync includes a status property.

Possible values: notStarted , running , succeeded , failed .

You must wait until the status is succeeded before processing results.

This ensures you don’t attempt to read lines before OCR has completed.

This corresponds to B.

Wrap in a loop with delay

Since OCR takes time, you need to poll the service until the status becomes succeeded or failed .

A typical pattern is:

ReadOperationResult results;

do

{

results = await client.GetReadResultAsync(Guid.Parse(operationId));

await Task.Delay( 1000 ); // wait 1 second before retrying

}

while (results.Status == OperationStatusCodes.Running || results.Status == OperationStatusCodes.NotStarted);

This corresponds to D.

Option Analysis

A. Remove the Guid.Parse(operationId) parameter

Incorrect. The operationId is required to fetch the OCR results.

B. Add code to verify the results.Status value

Correct. You must check whether the operation has completed.

C. Add code to verify the status of the txtHeaders.Status value

Incorrect. txtHeaders only contains metadata and the operation location. The OCR completion status comes from the results.Status in the GetReadResultAsync response, not from txtHeaders .

D. Wrap the call to GetReadResultAsync within a loop that contains a delay

Correct. This is how you poll until the operation has finished.

Correct Answer: B and D

Microsoft References

Azure Computer Vision Read API

Read text using the Computer Vision client library

Box 1: " categories " : [ " Locations " , " Persons " , " Organizations " ],

Locations, Persons, Organizations are in the outputs.

Scenario: Contoso plans to develop a searchable knowledgebase of all the intellectual property

Note: The categories parameter is an array of categories that should be extracted. Possible category types: " Person " , " Location " , " Organization " , " Quantity " , " Datetime " , " URL " , " Email " . If no category is provided, all types are returned.

Box 2: { " name " : " entities " }

The include wikis, so should include entities in the outputs.

Note: entities is an array of complex types that contains rich information about the entities extracted from text, with the following fields

name (the actual entity name. This represents a " normalized " form)

wikipediaId

wikipediaLanguage

wikipediaUrl (a link to Wikipedia page for the entity)

etc.

Within a search service, synonym maps are a global resource that associate equivalent terms, expanding the scope of a query without the user having to actually provide the term. For example, assuming " dog " , " canine " , and " puppy " are mapped synonyms, a query on " canine " will match on a document containing " dog " .

Create synonyms: A synonym map is an asset that can be created once and used by many indexes.

C: Analyze Receipt - Get Analyze Receipt Result.

Query the status and retrieve the result of an Analyze Receipt operation.

Request URL: https://{endpoint}/formrecognizer/v2.0-preview/prebuilt/receipt/analyzeResults/{resultId}

E: POST {Endpoint}/vision/v3.2/read/analyze

Use this interface to get the result of a Read operation, employing the state-of-the-art Optical Character Recognition (OCR) algorithms optimized for text-heavy documents.

Scenario: Contoso plans to develop a document processing workflow to extract information automatically from PDFs and images of financial documents

The document processing solution must be able to process standardized financial documents that have the following characteristics:

- Contain fewer than 20 pages.

- Be formatted as PDF or JPEG files.

- Have a distinct standard for each office.

*The document processing solution must be able to extract tables and text from the financial documents.

The document processing solution must be able to extract information from receipt images.

Can search content in different formats, including video

Audio and video insights (multi-channels). When indexing by one channel, partial result for those models will be available.

Keywords extraction: Extracts keywords from speech and visual text.

Named entities extraction: Extracts brands, locations, and people from speech and visual text via natural language processing (NLP).

Topic inference: Makes inference of main topics from transcripts. The 2nd-level IPTC taxonomy is included.

Artifacts: Extracts rich set of " next level of details " artifacts for each of the models.

Sentiment analysis: Identifies positive, negative, and neutral sentiments from speech and visual text.

Step 1: Create a new project and load sample documents

Create a new project. Projects store your configurations and settings.

Step 2: Label the sample documents

When you create or open a project, the main tag editor window opens.

Step 3: Train a custom model.

Finally, train a custom model.

Scenario: All planned projects must support English, French, and Portuguese.

If a bot uses multiple LUIS models and QnA Maker knowledge bases (knowledge bases), you can use the Dispatch tool to determine which LUIS model or QnA Maker knowledge base best matches the user input. The dispatch tool does this by creating a single LUIS app to route user input to the correct model.

The wiki contains text in English, French and Portuguese.

Scenario: All planned projects must support English, French, and Portuguese.

The Document Extraction skill extracts content from a file within the enrichment pipeline. This allows you to take advantage of the document extraction step that normally happens before the skillset execution with files that may be generated by other skills.

Note: The Translator Text API will be used to determine the from language. The Language detection skill is not required.

The requirements state:

All AI solution responses must have a confidence score ≥ 70%.

If the response confidence score is < 70%, the response must be improved by human input.

Members of the Consultant-Bookkeeper group must be able to process financial documents, which includes performing manual review when the AI confidence is below threshold.

Looking at the provided JSON:

" fields " : {

" ReceiptType " : {

" type " : " string " ,

" valueString " : " Itemized " ,

" confidence " : 0.672

},

" MerchantName " : {

" type " : " string " ,

" valueString " : " Tailwind " ,

" confidence " : 0.913

}

}

ReceiptType.confidence = 0.672 → this is below 0.7.

MerchantName.confidence = 0.913 → this is above 0.7.

Therefore, the correct condition for triggering manual review is:

documentResults.fields. < field > .confidence < 0.7

Now, evaluating the options:

A. documentResults.docType == " prebuilt:receipt "

Always true for receipts, does not check confidence. Not correct.

B. documentResults.fields.*.confidence < 0.7

This is the correct general expression: trigger manual review whenever any field confidence is below 0.7.

C. documentResults.fields.ReceiptType.confidence > 0.7

This would bypass manual review when ReceiptType has high confidence. The requirement is to trigger review when confidence < 0.7, so this is the opposite.

D. documentResults.fields.MerchantName.confidence < 0.7

This only checks one field (MerchantName). In the JSON, MerchantName has confidence 0.913 ( > 0.7), so this condition would not trigger, but ReceiptType clearly needs review. Too narrow.

Correct Answer: B. documentResults.fields.*.confidence < 0.7

Microsoft References

Azure AI Document Intelligence – Confidence scores

Human-in-the-loop for Document Intelligence

Box 1: Cognitive Service User

Ensure that the members of a group named Management-Accountants can approve the FAQs.

Approve=publish.

Cognitive Service User (read/write/publish): API permissions: All access to Cognitive Services resource except for ability to:

1. Add new members to roles.

2. Create new resources.

Box 2: Cognitive Services QnA Maker Editor

Ensure that the members of a group named Consultant-Accountants can create and amend the FAQs.

QnA Maker Editor: API permissions:

1. Create KB API

2. Update KB API

3. Replace KB API

4. Replace Alterations

5. " Train API " [in new service model v5]

Box 3: Cognitive Services QnA Maker Read

Ensure that the members of a group named the Agent-CustomerServices can browse the FAQs.

QnA Maker Read: API Permissions:

1. Download KB API

2. List KBs for user API

3. Get Knowledge base details

4. Download Alterations

Generate Answer

Scenario: When the response confidence score is low, ensure that the chatbot can provide other response options to the customers.

When no good match is found by the ranker, the confidence score of 0.0 or " None " is returned and the default response is " No good match found in the KB " . You can override this default response in the bot or application code calling the endpoint. Alternately, you can also set the override response in Azure and this changes the default for all knowledge bases deployed in a particular QnA Maker service.

Choosing Ranker type: By default, QnA Maker searches through questions and answers. If you want to search through questions only, to generate an answer, use the RankerType=QuestionOnly in the POST body of the GenerateAnswer request.

Scenario: Support autocompletion and autosuggestion based on all product name variants.

A: Call a suggester-enabled query, in the form of a Suggestion request or Autocomplete request, using an API. API usage is illustrated in the following call to the Autocomplete REST API.

POST /indexes/myxboxgames/docs/autocomplete?search & api-version=2020-06-30

{

" search " : " minecraf " ,

" suggesterName " : " sg "

}

B: In Azure Cognitive Search, typeahead or " search-as-you-type " is enabled through a suggester. A suggester provides a list of fields that undergo additional tokenization, generating prefix sequences to support matches on partial terms. For example, a suggester that includes a City field with a value for " Seattle " will have prefix combinations of " sea " , " seat " , " seatt " , and " seattl " to support typeahead.

F. Use the default standard Lucene analyzer ( " analyzer " : null) or a language analyzer (for example, " analyzer " : " en.Microsoft " ) on the field.

Box 1: Azure Blob storage

At the start of the pipeline, you have unstructured text or non-text content (such as images, scanned documents, or JPEG files). Data must exist in an Azure data storage service that can be accessed by an indexer.

Box 2: Computer Vision API

Scenario: Provide users with the ability to search insight gained from the images, manuals, and videos associated with the products.

The Computer Vision Read API is Azure ' s latest OCR technology (learn what ' s new) that extracts printed text (in several languages), handwritten text (English only), digits, and currency symbols from images and multi-page PDF documents.

Box 3: Translator API

Scenario: Product descriptions, transcripts, and all text must be available in English, Spanish, and Portuguese.

Box 4: Azure Files

Scenario: Store all raw insight data that was generated, so the data can be processed later.

Box 1: QnA Maker Editor

Scenario: Provide all employees with the ability to edit Q & As.

The QnA Maker Editor (read/write) has the following permissions:

Create KB API

Update KB API

Replace KB API

Replace Alterations

" Train API " [in

new service model v5]

Box 2: Contributor

Scenario: Only senior managers must be able to publish updates.

Contributor permission: All except ability to add new members to roles

Scenario: All videos must have transcripts that are associated to the video and included in product descriptions.

Product descriptions, transcripts, and all text must be available in English, Spanish, and Portuguese.

Step 1: Upload the video to blob storage

Given a video or audio file, the file is first dropped into a Blob Storage. T

Step 2: Index the video by using the Video Indexer API.

When a video is indexed, Video Indexer produces the JSON content that contains details of the specified video insights. The insights include: transcripts, OCRs, faces, topics, blocks, etc.

Step 3: Extract the transcript from the Video Indexer API.

Step 4: Translate the transcript by using the Translator API.

Parameter type: stream

Visual features list: VisualFeatureTypes.Description

Result to use: results.Description.Captions[0]

To generate accessible alt text , you should use the caption produced by Azure Computer Vision’s Description feature (it produces a human-readable sentence with a confidence score). Therefore:

The image input should be a stream , because you’re uploading images (not just passing URLs) during product creation and AnalyzeImage…Async supports image streams .

Request the Description feature in VisualFeatureTypes so the service returns results.Description.Captions.

Use results.Description.Captions[0] and return the caption text if its confidence is high enough (e.g., > 0.5) to meet the accessibility requirement that all images must have relevant alt text .

Other features (Tags, Objects, Brands) are useful for enrichment but do not directly return natural-language captions suitable for alt text.

Microsoft Azure AI Solution References

Computer Vision (Image Analysis) – Description/Captions and features: Microsoft Docs, Image Analysis – VisualFeatureTypes.Description returns description.captions.

https://learn.microsoft.com/azure/ai-services/computer-vision/concept-image-analysis

SDK usage (analyze image from a stream): Microsoft Docs, Analyze an image by using the Computer Vision client library .

https://learn.microsoft.com/azure/ai-services/computer-vision/how-to/call-analyze-image?tabs=version-3-2#analyze-an-image-from-a-stream

Comprehensive Detailed Explanation

The requirement is to build an Adaptive Card for the chatbot that supports multilingual product display and meets accessibility requirements.

Product Name Display

The product JSON stores localized names like " name " : { " en " : " ... " , " es " : " ... " , " pt " : " ... " } .

To dynamically render based on the current language, you must use name[language].

This ensures the product name shown in the card adapts automatically to the user’s preferred language.

Stock Level Warning

Business rules require warnings when stock is low or out of stock.

This means you display the stock warning only when stockLevel is not ' OK ' .

Therefore, the correct condition is:

" $when " : " ${stockLevel != ' OK ' } "

Image Alt Text for Accessibility

Accessibility requirements specify that all images must have alt text in English, Spanish, and Portuguese.

In the product JSON, " alttext " is also localized:

" altText " : { " en " : " Bicycle " , " es " : " Bicicleta " , " pt " : " Bicicleta " }

To render the correct localized alt text dynamically, use image.altText[language].

Correct Selections:

First blank: name[language]

Second blank: " $when " : " ${stockLevel != ' OK ' } "

Third blank: image.altText[language]

Microsoft References

Adaptive Cards Templating

Adaptive Card $when property

Multilingual data in Azure Cognitive Search and Adaptive Cards

Box 1: api-nam.cognitive.microsofttranslator.com

https://docs.microsoft.com/en-us/azure/cognitive-services/translator/reference/v3-0-reference

Box 2: /translate