✅ AI-102 Complete SDK Summary

(Computer Vision, Speech, Language, Conversational AI, Form Recognizer, Cognitive Search, Bot Framework SDK)

🚀 1. Computer Vision SDK (Azure.AI.Vision / Azure.CognitiveServices.Vision)

Core client

var client = new ComputerVisionClient(
    new ApiKeyServiceClientCredentials(key))
{
    Endpoint = endpoint
};

1.1 Image Tagging

Exam pattern: TagImageAsync / TagImageInStreamAsync

using (var stream = File.OpenRead(path))
{
    var result = await client.TagImageInStreamAsync(stream);
    foreach (var tag in result.Tags)
        Console.WriteLine($"{tag.Name} {tag.Confidence}");
}

✔ Frequently asked:

Tags = simple labels
Not for face recognition
Works with file streams

1.2 Smart Cropping / Thumbnail

API call:

POST https://{resource}.cognitiveservices.azure.com/vision/v3.1/generateThumbnail

1.3 OCR & Read API (ASYNC LOOP PATTERN)

This appears many times on exam questions.

var headers = await client.ReadAsync(stream);
string id = headers.OperationLocation.Split('/').Last();

ReadOperationResult result;
do
{
    result = await client.GetReadResultAsync(Guid.Parse(id));
    await Task.Delay(1000);
}
while (result.Status == OperationStatusCodes.Running
    || result.Status == OperationStatusCodes.NotStarted);

✔ Must: loop + delay ✔ Must: check result.Status

🚀 2. Speech SDK

Core setup

var speechConfig = SpeechConfig.FromSubscription(key, region);

2.1 Speech-to-Text

var recognizer = new SpeechRecognizer(speechConfig);
var result = await recognizer.RecognizeOnceAsync();
Console.WriteLine(result.Text);

Exam patterns:

Use SpeechRecognizer for STT
Not KeywordRecognizer, SpeakerRecognizer, or SpeechSynthesizer

2.2 Text-to-Speech

var synthesizer = new SpeechSynthesizer(speechConfig);
await synthesizer.SpeakTextAsync("Hello world");

2.3 Speech Translation

speechConfig.SpeechRecognitionLanguage = "fr";

var translator = new TranslationRecognizer(speechConfig);
translator.AddTargetLanguage("de");

2.4 Streaming MP3 Speech Input

var audioFormat = AudioStreamFormat.GetCompressedFormat(AudioStreamContainerFormat.MP3);
var audioConfig = AudioConfig.FromStreamInput(pushStream, audioFormat);

var recognizer = new SpeechRecognizer(speechConfig, audioConfig);

✔ MP3 requires GetCompressedFormat ✔ Streaming uses push stream

2.5 Custom Neural Voice

Exam pattern:

Create voice in Speech Studio portal
Generate narration with Text-to-Speech service

🚀 3. Form Recognizer / Document Intelligence SDK

Analyze PDFs, images

You use DocumentAnalysisClient

var client = new DocumentAnalysisClient(new Uri(endpoint), new AzureKeyCredential(key));
var result = await client.AnalyzeDocumentAsync("prebuilt-invoice", stream);

Exam facts:

Use Custom Neural model for handwritten/custom forms
Use prebuilt models (invoice, receipt, ID) for minimal development
Labeling tool:
create project
upload samples
label documents
train model

🚀 4. Azure AI Language Service SDKs

4.1 Text Analytics (Sentiment, NER, PII)

Package:

Python → azure-ai-textanalytics
C# → Azure.AI.TextAnalytics

Disable logging (VERY common exam question):

?loggingOptOut=true

4.2 Conversational Language Understanding (CLU)

For intent recognition:

SDK packages:

C#: Azure.AI.Language.Conversations
Python: azure-ai-language-conversations

4.3 Language Detection

Endpoint:

https://<region>.api.cognitive.microsoft.com/text/analytics/v3.1/languages

4.4 Active Learning (LUIS Legacy)

Enable logging:

log=true

✔ Required for review endpoint utterances ✔ Not related to show-all-intents or sentiment

4.5 Phrase Lists

await client.Features.AddPhraseListAsync(appId, versionId,
    new PhraselistCreateObject {
        Name="PL1", Phrases="item1,item2"
    });

4.6 Handle "None" Intent

To reduce misclassification:

Add examples to None intent

🚀 5. Azure Cognitive Search SDK

5.1 Field attributes

Common exam case:

Requirement	Attribute
Search text	`searchable`
Return field	`retrievable`
Filter/drill-down	`facetable`
Filter-only	`filterable`
Sort	`sortable`

5.2 Use in indexers with Form Recognizer

To export to Power BI: ✔ Use table projection

5.3 Throttling

✔ Add replicas to handle queries ✔ Add partitions to handle ingestion

🚀 6. Bot Framework SDK (C# or Python)

State Access

_userState.CreateProperty<UserProfile>("UserProfile");
_conversationState.CreateProperty<ConversationData>("ConversationData");

Memory storage:

Yes stores state
No persistence after restart

6.1 QnA Maker / Question Answering

To query a KB: ✔ Use QnAMakerDialog

6.2 Reset conversation after inactivity

Python exam snippet:

await turn_context.send_activity("Welcome back! Let's start over...")
await self.conversation_state.clear_state(turn_context)

6.3 Bot with Speech

To enable voice in Composer:

✔ Configure Speech language/voice ✔ Add Speech resource key/endpoint ✔ Add Speech to responses

Correct answers: A, B, E

🚀 7. Containers (LUIS / Cognitive Services)

Steps to run a LUIS container:

Select latest deployable version
Export for container (GZIP)
Run Docker container and mount model folder

Container basics:

/status → API key validity
/swagger → API docs

🎯 Final: What to Memorize for the Exam

Read API requires polling loop
Speech translation: SpeechRecognitionLanguage + AddTargetLanguage()
Logging for active learning: log=true
Disable storage: loggingOptOut=true
Cognitive Search field mapping
CLU models → Conversations SDK package
Form Recognizer → Custom Neural for handwriting
QnA Maker queries use QnAMakerDialog

Here are your AI-102 SDK Flashcards — short, exam-focused, and perfect for fast memorization. If you want, I can turn these into Anki deck, PDF, or printable cards.

🎴 AI-102 SDK Flashcards

(Organized by service; each section ~10–15 cards)

🟥 Computer Vision SDK

CV-1 — Create Vision Client

Q: How do you create a ComputerVisionClient in C#? A:

var client = new ComputerVisionClient(
    new ApiKeyServiceClientCredentials(key))
{
    Endpoint = endpoint
};

CV-2 — Generate Image Description

Q: Which method returns full-sentence descriptions? A: DescribeImageInStreamAsync()

CV-3 — Tag Image

Q: Which method returns only tags? A: TagImageInStreamAsync()

CV-4 — Read OCR Pattern

Q: What pattern is required to use Read API? A:

Call ReadAsync()
Extract Operation ID
Poll with GetReadResultAsync() until Succeeded

CV-5 — Read Operation IDs

Q: Where is the Read API operation ID found? A: In the OperationLocation response header.

CV-6 — OCR Result Status Values

Q: Which statuses require polling? A: Running and NotStarted

CV-7 — Smart Thumbnail

Q: Which API generates smart crops? A: /generateThumbnail

CV-8 — Model Domain Analysis

Q: Which method uses domain-specific models (e.g., celebrities)? A: AnalyzeImageByDomainInStreamAsync()

🟦 Speech SDK

SP-1 — Initialize Speech Config

Q: How to initialize SpeechConfig? A:

var config = SpeechConfig.FromSubscription(key, region);

SP-2 — STT (Speech-to-Text) Basic

Q: Which class is used for speech recognition? A: SpeechRecognizer

SP-3 — Recognize Once

Q: Which method returns a single utterance? A: RecognizeOnceAsync()

SP-4 — TTS (Text-to-speech)

Q: Which class performs TTS? A: SpeechSynthesizer

SP-5 — Streaming Audio (MP3)

Q: How to stream MP3 audio into Speech SDK? A: Use AudioConfig.FromStreamInput() with AudioStreamFormat.GetCompressedFormat(AudioStreamContainerFormat.MP3)

SP-6 — Speech Translation

Q: How to add target translation languages? A: translator.AddTargetLanguage("de")

SP-7 — Custom Neural Voice

Q: Where is custom voice model created? A: Speech Studio → Custom Neural Voice

🟩 Form Recognizer (Document Intelligence)

FR-1 — Create Client

Q: What is the main client class? A: DocumentAnalysisClient

FR-2 — Prebuilt Model Example

Q: How to analyze an invoice? A:

client.AnalyzeDocumentAsync("prebuilt-invoice", stream)

FR-3 — Custom Model Training Steps

Q: Steps to create custom model? A:

Upload documents
Label fields
Train model
Analyze documents

FR-4 — Handwriting / Complex Layout

Q: Which model type handles noisy layouts or handwriting? A: Custom Neural model

FR-5 — Azure Storage Requirement

Q: Does training require storage account? A: Yes — sample docs stored in a container.

🟨 Language (Text Analytics + CLU + PII)

TA-1 — Client Initialization

Q: Which client is used for Text Analytics? A: TextAnalyticsClient

TA-2 — Disable Logging

Q: How to disable logging for privacy? A: Add ?loggingOptOut=true to your endpoint.

TA-3 — Sentiment Analysis

Q: Which method performs sentiment analysis? A: AnalyzeSentimentAsync()

TA-4 — Language Detection endpoint

Q: What is the language detection endpoint path? A: /text/analytics/v3.1/languages

TA-5 — PII Recognition

Q: Which method detects PII entities? A: RecognizePiiEntitiesAsync()

CLU-1 — CLU Client

Q: What is the CLU SDK package? A: Azure.AI.Language.Conversations

CLU-2 — Misclassified "None" Intent

Q: How to reduce misclassification into "None"? A: Add more examples to the None intent.

LUIS-1 — Enable Active Learning

Q: How to collect utterances for review? A: Use query parameter log=true.

🟫 Cognitive Search SDK

CS-1 — Search Index Client

Q: Which client uploads documents? A: SearchClient (or SearchIndexClient for schema)

CS-2 — Field Attributes

Q: What does searchable mean? A: Full-text search enabled.

CS-3 — Facetable

Q: What is the field attribute used for aggregations / drill-down? A: facetable

CS-4 — Filterable

Q: What is filterable used for? A: OData filters (no text analysis)

CS-5 — Sorting

Q: Which attribute must be enabled to sort on a field? A: sortable

CS-6 — High throughput ingestion

Q: For heavy indexing, what should be increased? A: Partitions

CS-7 — High query load

Q: For many queries, what should be increased? A: Replicas

CS-8 — Indexer + Form Recognizer

Q: What projection is required for Power BI export? A: Table projection

🟧 Bot Framework SDK

BOT-1 — State Storage

Q: Which classes manage user and conversation state? A:

UserState
ConversationState

BOT-2 — Clear Conversation

Q: How to reset bot conversation? A: conversationState.clear_state(turn_context)

BOT-3 — QnA Maker / Language Question Answering

Q: Which dialog queries the KB? A: QnAMakerDialog

BOT-4 — Memory Storage Behavior

Q: Does MemoryStorage persist between restarts? A: No — volatile only.

BOT-5 — Add speech to bot

Q: What must be configured for voice bots? A:

Speech key
Speech region
Output voice
Set response as speech-enabled

🟪 Containers (Vision, Language, LUIS, etc.)

CT-1 — Deployable LUIS version

Q: Which LUIS versions can run in a container? A: Only deployable versions.

CT-2 — LUIS container deployment steps

Q: Steps? A:

Export container package (GZIP)
Run Docker container
Mount model folder
Call container endpoint

CT-3 — Container Test URLs

Q: What does /status return? A: API key validation.

CT-4 — Swagger

Q: Which endpoint exposes API documentation? A: /swagger

❗ Trap 1 — Wrong Computer Vision method

If question says “full sentence for visually impaired” → DescribeImageInStreamAsync()

❗ Trap 2 — Read API requires polling

Never pick "immediate result" methods.

❗ Trap 3 — Cognitive Search load

Ingestion → partitions
Queries → replicas

❗ Trap 4 — CLU vs. Text Analytics

Intent classification → CLU
Entity extraction → Text Analytics

❗ Trap 5 — Logging for Active Learning

Enable: log=true
Disable: loggingOptOut=true

Great — below is a 100-question, dev-focused AI-102 SDK practice exam. Questions focus on code, SDK classes/methods, call patterns, and common exam traps across Computer Vision, Speech, Form Recognizer / Document Intelligence, Language/CLU, Cognitive Search, Bot Framework, and Containers.

I kept questions short and tightly targeted. After the 100 questions you’ll find a compact answer key with one-line rationales so you can self-grade quickly.

AI-102 — 100 Dev-Focused SDK Questions

(Each question is multiple choice: A, B, C, or D.)

1

You want a full-sentence description for an image to present to visually impaired users. Which method should you call?

A. TagImageInStreamAsync
B. DescribeImageInStreamAsync
C. ReadInStreamAsync
D. AnalyzeImageByDomainInStreamAsync

2

You need to poll the Computer Vision Read API until OCR completes. Which sequence is correct?

A. ReadAsync → GetReadResultAsync with operation ID loop
B. GetReadResultAsync → ReadAsync
C. ReadInStreamAsync → DescribeImageInStreamAsync
D. TagImageInStreamAsync → GetReadResultAsync

3

Which class is used to perform speech-to-text in the Speech SDK (C#)?

A. SpeechSynthesizer
B. KeywordRecognizer
C. SpeechRecognizer
D. SpeakerRecognizer

4

To stream MP3-encoded audio into the Speech SDK push stream, which method sets the format?

A. AudioStreamFormat.GetWaveFormatPCM()
B. AudioStreamFormat.GetCompressedFormat(AudioStreamContainerFormat.MP3)
C. AudioStreamFormat.CreateDefault()
D. AudioStreamFormat.GetFloat32Format()

5

Which method produces tags (keywords) for an image (no sentences)?

A. DescribeImageInStreamAsync
B. ReadInStreamAsync
C. TagImageInStreamAsync
D. AnalyzeImagesByDomainInStreamAsync

6

You must disable logging/persistence of user text to the Language service. Which query parameter is used?

A. model-version
B. log=true
C. loggingOptOut=true
D. showStats=true

7

Which SDK package provides conversational intent recognition support in C#?

A. Azure.AI.Language.Conversations
B. Azure.AI.TextAnalytics
C. Microsoft.Cognitive.Speech
D. Azure.CognitiveServices.Vision

8

You need to extract structured fields from receipts using the prebuilt Form Recognizer model with the SDK. Which client+method is correct?

A. FormTrainingClient + StartRecognizeReceiptsFromUri
B. FormRecognizerClient + StartRecognizeReceiptsFromUri
C. DocumentAnalysisClient + AnalyzeDocumentAsync("prebuilt-receipt", uri)
D. ComputerVisionClient + ReadAsync

9

Which attribute must be set on a Cognitive Search field to allow drill-down filters (facets)?

A. searchable
B. retrievable
C. facetable
D. filterable

10

To increase query throughput for Azure Cognitive Search, you should:

A. add partitions
B. add replicas
C. add indexes
D. enable CMK encryption

11

You want to synthesize speech (text-to-speech). Which class do you instantiate?

A. SpeechRecognizer
B. SpeechSynthesizer
C. TranslationRecognizer
D. SpeechClient

12

Which code fragment correctly polls a Read OCR operation?

A. await client.ReadAsync(stream); results = await client.GetReadResultAsync(Guid.Parse(id)); // no loop
B. Do { results = await client.GetReadResultAsync(Guid.Parse(id)); await Task.Delay(1000); } while (status==Running)
C. results = await client.ReadAsync(stream); // no follow-up
D. await client.GetReadResultAsync(Guid.NewGuid()); // random id

13

Which Speech SDK object would you use to perform real-time translation and add target languages?

A. SpeechRecognizer
B. TranslationRecognizer
C. SpeechSynthesizer
D. AudioConfig

14

When building a bot that queries a Q&A knowledge base, which dialog class is most appropriate?

A. ComponentDialog
B. AdaptiveDialog
C. QnAMakerDialog
D. SkillDialog

15

You need to collect utterances for active learning so reviewers can improve the model. Which query parameter must you include when calling the prediction endpoint?

A. show-all-intents=true
B. log=true
C. loggingOptOut=true
D. verbose=false

16

Which client class is appropriate to call prebuilt Document Intelligence analysis in new SDKs?

A. DocumentAnalysisClient
B. FormRecognizerClient only
C. DocumentClient
D. AnalysisClient

17

You must stream MP3 audio input and use SpeechRecognizer with a push stream. Which AudioConfig is correct?

A. AudioConfig.FromDefaultMicrophoneInput()
B. AudioConfig.FromStreamInput(pushStream, AudioStreamFormat.GetCompressedFormat(...))
C. AudioConfig.FromWavFileInput(filePath)
D. AudioConfig.FromCustomStream(pushStream) // no format

18

To call a domain-specific Vision model for landmark detection, which method is appropriate?

A. DescribeImageInStreamAsync
B. AnalyzeImagesByDomainInStreamAsync
C. TagImageInStreamAsync
D. ReadInStreamAsync

19

You want to create a new CLU conversational client in Python to call intent model. Which package would you install?

A. azure-ai-language-conversations
B. azure-cognitiveservices-language-luis
C. azure-cognitiveservices-speech
D. azure-ai-textanalytics

20

Which Cognitive Search field attribute allows full-text search inside a field?

A. filterable
B. retrievable
C. searchable
D. facetable

21

Which method is used to generate a smart-cropped thumbnail (resized image) using Computer Vision? A. generateThumbnail endpoint B. DescribeImage endpoint C. TagImage endpoint D. Read API

22

Which Bot Framework state storage is non-persistent and loses data on restart?

A. BlobStorage
B. CosmosDbStorage
C. MemoryStorage
D. AzureTableStorage

23

Which command-line behavior should you avoid when running Cognitive Services containers to prevent keys being saved in command history?

A. Passing keys via environment variables in a script loaded at runtime
B. Passing keys directly in docker run command line
C. Using Azure Key Vault and mounting secrets at runtime
D. Using a docker run script that reads keys from protected file

24

You exported a LUIS app for container use. Which file format should you export to run in a container?

A. .json (plain)
B. .zip with model files
C. Export for container (GZIP)
D. .yaml

25

Which Form Recognizer method returns general layout (lines, tables) rather than receipts-specific fields?

A. StartRecognizeReceiptsFromUri
B. StartRecognizeContentFromUri
C. AnalyzeDocumentAsync("prebuilt-invoice")
D. ReadAsync

26

Which SDK parameter disables service-side logging for PII-sensitive calls? A. piiCategories B. loggingOptOut=true C. showStats=false D. model-version=latest

27

What is the typical response when you call the Computer Vision Read API immediately (before completion)?

A. Final text results
B. OperationLocation header with operation ID
C. 404 Not Found
D. Tag list

28

Which method should you call to add a phrase list to a LUIS model programmatically (per the book snippet)? A. client.Features.AddPhraseListAsync(...) B. client.Models.AddPhrases(...) C. client.PhraseLists.Create(...) D. client.AddPhrasesToModel(...)

29

Which Speech SDK method performs a single utterance synchronous recognition?

A. StartContinuousRecognitionAsync
B. RecognizeOnceAsync
C. RecognizeAsyncMultiple
D. GetRecognitionResult

30

Which field attributes must you set so a field appears in results and is searchable and can be used for facets?

A. retrievable, filterable, sortable
B. searchable, facetable, retrievable
C. retrievable, facetable, key
D. searchable, retrievable, sortable

31

You want to use the prebuilt Text Analytics language detection endpoint. Which path is correct? A. /text/analytics/v3.1/languages B. /language/detect/v1 C. /translate?api-version=3.0 D. /language/identify

32

Which class is used to train custom Form Recognizer models (older SDK)?

A. FormTrainingClient
B. FormRecognizerClient
C. DocumentTrainingClient
D. TrainingClient

33

You want translation in a speech app (speech in → speech out). Which combination best fits minimal work?

A. SpeechRecognizer + custom model + SpeechSynthesizer
B. TranslationRecognizer + SpeechSynthesizer (or built-in TTS)
C. SpeechRecognizer + Text Analytics + SpeechSynthesizer
D. KeywordRecognizer + Translation API

34

Which Bot Framework method resets conversation state in Python?

A. conversationState.delete(turn_context)
B. conversationState.clear_state(turn_context)
C. bot.clear()
D. conversationState.reset()

35

Which attribute in an Azure Cognitive Search index is required to allow sorting on a field?

A. facetable
B. retrievable
C. sortable
D. searchable

36

Which Speech SDK method is used to synthesize SSML or text to speaker?

A. recognizer.SpeakAsync()
B. synthesizer.SpeakSsmlAsync or SpeakTextAsync
C. translator.SpeakAsync()
D. audioPlayer.PlayText()

37

Which Computer Vision method returns domain-specific outputs like celebrity names?

A. TagImageInStreamAsync
B. DescribeImageInStreamAsync
C. AnalyzeImagesByDomainInStreamAsync
D. ReadInStreamAsync

38

Which SDK client is used to call Text Analytics sentiment analysis?

A. TextAnalyticsClient
B. LanguageClient
C. SentimentClient
D. AITextClient

39

Which property must you check in Read/GetReadResult responses to know if status is still running?

A. results.State
B. results.Status
C. results.IsComplete
D. results.Progress

40

Which approach allows only VNet resources to access a Language service (ta1) — per exam patterns?

A. Configure NSG on vnet1
B. Configure virtual network (private endpoint) settings on ta1
C. Deploy language service container in vnet1 only
D. Use Azure Firewall for vnet1

41

Which SDK method is appropriate to extract receipts from a remote URI using prebuilt model (per snippet)?

A. FormRecognizerClient.StartRecognizeReceiptsFromUri(new Uri(...))
B. DocumentAnalysisClient.AnalyzeDocumentAsync("prebuilt-receipt", new Uri(...))
C. ComputerVisionClient.ReadAsync(new Uri(...))
D. TextAnalyticsClient.RecognizeEntitiesFromUri(...)

42

Which Bot Framework dialog should be used to route user queries across multiple skills/recognizers (Orchestrator-like)?

A. QnAMakerDialog
B. AdaptiveDialog
C. Orchestrator runtime component (not a dialog)
D. ComponentDialog

43

When running a Cognitive Services container, which local endpoint helps you verify API key validity?

A. /health
B. /status
C. /validateKey
D. /keycheck

44

Which service is best for detecting anomalies across many correlated IoT sensor time series with root cause analysis and alerting out-of-the-box?

A. Anomaly Detector
B. Metrics Advisor (Azure Metrics Advisor)
C. Azure Machine Learning
D. Custom Vision

45

Which Search indexer projection should you use if you want Power BI to work with results easily?

A. object projection
B. file projection
C. table projection
D. projection group

46

What must you configure in a bot to support speech channels using Bot Framework Composer? (Choose the trio that belongs together in exam-style)

A. Add Orchestrator, Add endpoint key, Add App Insights
B. Configure Speech language/voice, Add endpoint & key of Speech resource, Add speech to bot responses
C. Add LUIS, Add QnA Maker, Add Direct Line
D. Add MemoryStorage, Add BlobStorage, Add CosmosDB

47

Which method returns the operation ID location for async Read OCR?

A. response.Headers.OperationLocation
B. response.Content.OperationId
C. response.Json.operation_id
D. response.Headers.CorrelationId

48

You must avoid logging API keys or billing info into shell history when launching containers. Best practice:

A. Hardcode keys in scripts and run them once.
B. Provide keys as command-line args to docker run.
C. Use environment variables injected from a secure store or mount files with restricted permissions.
D. Paste keys interactively into terminal before running container.

49

Which class (package) is used in C# to call Language Conversations models (per earlier snippet)?

A. Azure.AI.Language.Conversations (package)
B. Microsoft.Cognitive.LUIS.Client
C. Azure.CognitiveServices.LanguageClient
D. Azure.AI.LanguageSDK

50

Which method would you use to analyze layout (tables, lines) rather than extracting semantic key/value pairs?

A. StartRecognizeReceiptsFromUri
B. StartRecognizeContentFromUri / AnalyzeLayout equivalent
C. AnalyzeDocumentAsync("prebuilt-receipt")
D. DescribeImageInStreamAsync

51

Which of these is the correct pattern to add a phrase list to many LUIS models programmatically (based on the book)?

A. client.Features.AddPhraseListAsync(appId, versionId, new PhraselistCreateObject{...})
B. client.Models.AddPhraseList(appId, phrases)
C. client.AddPhrases(appId, new PhraseObject(phrases))
D. client.CreatePhraseList(appId, phrases)

52

Which Text Analytics method detects PII entities?

A. AnalyzeSentimentAsync
B. RecognizePiiEntitiesAsync
C. ExtractKeyPhrasesAsync
D. DetectLanguageAsync

53

Which Speech SDK flow is appropriate for long-running continuous recognition from microphone?

A. RecognizeOnceAsync
B. StartContinuousRecognitionAsync / StopContinuousRecognitionAsync
C. ContinuousRecognizeOnce
D. BeginContinuousTranscribe

54

Which Computer Vision SDK output type helps visually impaired users by providing complete-sentence descriptions?

A. Tag list
B. Read text only
C. Descriptions (full sentences)
D. Domain model metadata

55

You want to use the latest deployable LUIS version for a container. Which is correct sequence?

A. Select latest version → Export for container (GZIP) → Run container and mount model
B. Train locally → Run container → Upload model
C. Export JSON → Import into container → Train in container
D. Run container → Publish version → Export

56

Which Bot Framework pattern will display a QnA answer only when score >= threshold? (Config in QnAMakerOptions)

A. QnAMakerOptions.ScoreThreshold
B. QnAMakerOptions.MaxAnswers
C. QnAMakerOptions.RankerType
D. QnAMakerOptions.CardNoMatchText

57

Which Document Intelligence resource name type should you create if you want to use prebuilt invoice analysis and download JSON output?

A. ComputerVision resource
B. Document Intelligence (Form Recognizer) resource
C. Text Analytics resource
D. Custom Vision resource

58

Which Azure Search change will NOT help reduce throttling for query load?

A. Add replicas
B. Move to higher tier
C. Add indexes
D. Increase number of replicas

59

Which SDK concept do you use to avoid persisting sensitive user text in Language service logs?

A. set header X-Privacy: true
B. append ?disableLogging=true
C. use query parameter loggingOptOut=true
D. set model-version=0

60

Which endpoint is used to detect language with Text Analytics v3.1 (per book snippet)? A. https://.api.cognitive.microsoft.com/text/analytics/v3.1/languages B. https://.api.cognitive.microsoft.com/language/detect C. https://.api.cognitive.microsoft.com/translate/detect D. https://.api.cognitive.microsoft.com/text/language

61

You need to run a Text Analytics Sentiment Analysis container on VM with Docker. Which docker image name is correct (per snippet)?

A. mcr.microsoft.com/azure-cognitive-services/textanalytics/sentiment
B. microsoft/textanalytics:sentiment
C. azure/textanalytics/sentiment:latest
D. mcr.microsoft.com/cognitive/text/sentiment

62

Which Form Recognizer client method should you use to analyze receipts from a URI and wait for completion (older SDK snippet)?

A. StartRecognizeReceiptsFromUri(...).WaitForCompletionAsync()
B. StartAnalyzeReceipts(new Uri(...))
C. AnalyzeDocumentAsync("prebuilt-receipt") // no wait
D. RecognizeContentAsync(uri)

63

Which Bot Framework storage accessor creation syntax is correct in C#?

A. _userState.CreateProperty<UserProfile>(nameof(UserProfile));
B. _userState.GetProperty<UserProfile>("UserProfile");
C. _conversationState.CreateAccessor<UserProfile>();
D. UserState.CreateProperty<UserProfile>("UserProfile"); // static

64

Which Speech SDK action is required to use Custom Neural Voice generation?

A. Create voice model in Speech Studio and then generate using Text-to-Speech API
B. Train voice model with SpeechSynthesizer locally
C. Use TTS with a special SDK flag UseCustomVoice=true only
D. No special steps needed; just pass voice name

65

Which of the following is NOT a correct reason to attach an S0 Cognitive Services resource to a Search skillset when processing 50k scanned docs?

A. F0 (free) tier has too few enrichments → choose S0
B. S0 supports scale for many documents
C. S0 eliminates need for Form Recognizer entirely
D. Free tier would quickly exceed limits

66

Which property of Search index fields increases index size and slightly increases query times when using CMK?

A. adding more replicas
B. enabling encryption (CMK)
C. setting searchable=true
D. adding facetable fields

67

Which SDK/class would you use to call Azure Video Analyzer (Video Indexer) insights programmatically?

A. VideoIndexer REST APIs (or SDK)
B. ComputerVisionClient
C. FormRecognizerClient
D. SpeechRecognizer

68

Which method would you use to get the operation status for a Read request in Computer Vision?

A. GetReadResultAsync(Guid)
B. GetReadStatus()
C. CheckReadOperation()
D. ReadStatusAsync()

69

Which package should be added to a C# app to call conversational Language service model (Model1) per earlier snippet?

A. Azure.AI.Language.Conversations
B. azure-cognitiveservices-language-luis
C. Universal.Microsoft.CognitiveServices.Speech
D. Xamarin.Cognitive.Speech

70

Which SDK method is used to start a recognition of general content (layout) from a URI (older Form Recognizer naming)?

A. StartRecognizeContentFromUri
B. StartRecognizeReceiptsFromUri
C. StartRecognizeLayoutFromUri
D. AnalyzeDocumentFromUri

71

Which setting in QnAMakerOptions controls minimum confidence to return an answer?

A. RankerType
B. ScoreThreshold
C. Top = 1
D. ConfidenceMin

72

Which is the correct order to move a Language (QnA) project to a different region?

A. Export project → Import into new Language service → Train & Publish
B. Recreate KB manually → Train → Publish
C. Train in new region → Export from old → Import old
D. Export → Publish → Import

73

When calling the Speech SDK for streaming MP3 audio, what is essential to include?

A. set SpeechRecognitionLanguage="en-US" only
B. use AudioStreamFormat.GetCompressedFormat(AudioStreamContainerFormat.MP3) for audio config
C. use AudioConfig.FromWavFileInput with mp3 path
D. use default audio config and set Format="MP3"

74

Which of these is required to allow only VNet1 to access a Cognitive Services resource?

A. configure the Cognitive Services resource virtual network / private endpoint
B. set NSG on VNet1 only
C. add firewall rules only
D. deploy service to public endpoint and add IP restrictions

75

Which method is used to analyze a remote image using the Describe API (older SDK)?

A. client.DescribeImageInStreamAsync(stream)
B. client.DescribeImageAsync(imageUrl)
C. client.AnalyzeImageByDomainInStreamAsync(stream)
D. client.TagImageAsync(imageUrl)

76

Which attribute must be true for an Azure Search field to be returned in search results?

A. retrievable = true
B. searchable = true
C. facetable = true
D. filterable = true

77

Which of the following is NOT true about MemoryStorage in Bot Framework?

A. It's in-memory only and non-persistent.
B. It persists data across app restarts.
C. Good for local testing.
D. Lost when process restarts.

78

You need to ensure your bot's QnA responses are only shown when confidence >= 0.6. Which option do you set?

A. QnAMakerOptions.ScoreThreshold = 0.6F
B. QnAMakerOptions.MaxAnswers = 0.6F
C. QnAMakerOptions.RankerType = "High"
D. QnAMakerOptions.CardNoMatchText = "No match"

79

Which Azure SDK client is used for general Text Analytics (sentiment, key phrases)?

A. TextAnalyticsClient
B. LanguageClient
C. SentimentClient
D. AnalyzeTextClient

80

Which of the following best practice avoids exposing keys in terminal history when running docker run?

A. use environment variables read from a file with 600 permissions and --env-file
B. pass -e "API_KEY=..." on command line
C. echo key into terminal then run docker run in same line
D. use history -c afterwards

81

Which of the following identifies the correct method to run sentiment analysis on text?

A. textClient.AnalyzeSentimentAsync(document)
B. textClient.RecognizePiiEntitiesAsync(document)
C. textClient.DetectLanguageAsync(document)
D. textClient.ExtractKeyPhrasesAsync(document)

82

Which Cognitive Search change increases the capacity for indexing (ingestion) operations?

A. Add replicas
B. Add partitions
C. Add more indexes
D. Enable CMK

83

Which SDK method would you use to start recognition of content (layout) locally with file stream?

A. client.StartRecognizeContentFromStream(stream)
B. client.ReadAsync(stream)
C. client.StartAnalyzeLayoutFromStream(stream)
D. client.StartRecognizeReceiptsFromStream(stream)

84

Which method is typically used to regenerate the secondary subscription key of a Cognitive Services account via management REST API?

A. POST .../regenerateKey with body {"keyName":"Key2"}
B. PUT .../keys/rotate secondary
C. GET .../keys/regenerate
D. PATCH .../keys with "rotate"

85

Which of these is the correct way to create SpeechConfig in C#?

A. var speechConfig = SpeechConfig.FromSubscription("key", "region");
B. var speechConfig = new SpeechConfig("key", "region");
C. var speechConfig = SpeechConfig.Create("key");
D. var speechConfig = SpeechConfig.FromKeyVault(kvUri);

86

Which of these is correct for using TranslationRecognizer to translate from French to German?

A. speechConfig.SpeechRecognitionLanguage = "fr"; translator.AddTargetLanguage("de");
B. speechConfig.SpeechRecognitionLanguage = "de"; translator.AddTargetLanguage("fr");
C. speechConfig.SpeechRecognitionLanguage = "fr"; translator.AddSourceLanguage("de");
D. translator.SetSource("fr"); translator.AddTarget("de"); // not SDK

87

Which resource should you attach to a Search skillset to perform OCR and text analytics at scale?

A. free Cognitive Services (F0)
B. S0 Cognitive Services resource
C. Computer Vision resource only
D. No resource required

88

Which class do you use to call AnalyzeDocumentAsync for prebuilt invoice in new SDK?

A. DocumentAnalysisClient
B. FormRecognizerClient
C. InvoiceAnalysisClient
D. DocumentIntelligenceClient

89

Which method on SpeechRecognizer returns the recognized text result for a single utterance?

A. RecognizeOnceAsync()
B. GetResultAsync()
C. StartSingleRecognition()
D. TranscribeAsync()

90

When processing receipts with Form Recognizer prebuilt model, which fields are commonly returned?

A. merchantName, transactionTotal, transactionDate
B. descriptionOnly
C. imageColors
D. facialAttributes

91

Which of these endpoints should you call to check swagger docs of a Cognitive Services container?

A. /swagger
B. /docs
C. /api-docs
D. /help

92

Which is the correct practice for programmatically creating a free Computer Vision resource via a helper method create_resource? Example from book: create_resource(client, "res1", "ComputerVision", "FO", "westus"). What does "FO" indicate?

A. Free tier (F0)
B. Standard tier
C. Billing region
D. Resource group

93

Which is the correct SDK client for analyzing video insights from Azure Video Indexer?

A. Use Video Indexer REST APIs / SDK
B. ComputerVisionClient.AnalyzeVideo
C. FormRecognizerClient.AnalyzeVideo
D. VideoClient from Speech SDK

94

Which of the following is true about QnA Maker provisioning in a resource group with an App Service plan? (Per book)

A. Azure Cognitive Search and an App Service instance are automatically created
B. LUIS authoring is automatically created
C. Azure SQL DB is always created
D. Azure Storage is never used

95

Which Speech SDK method provides continuous recognition for long audio files?

A. RecognizeOnceAsync
B. StartContinuousRecognitionAsync
C. TranscribeFileOnce
D. GetLongRecognition

96

Which of these is the correct approach to avoid storing billing/API keys in container command history?

A. Use an docker run script that loads secrets from a file not in shell history
B. Always paste the key into terminal directly
C. Hardcode key in the docker image
D. Print keys as part of container logs

97

Which attribute must be set for a Search field to be used for OData filter expressions?

A. facetable
B. filterable
C. searchable
D. retrievable

98

Which SDK action logs endpoint utterances for LUIS active learning? A. add header X-Active-Learning: true B. include log=true in predict request C. call /logUtterance endpoint D. set verbose=true only

99

Which class is more appropriate to call Text Analytics PII detection in C#?

A. Azure.AI.TextAnalytics.TextAnalyticsClient
B. Azure.AI.Language.Conversations.ConversationsClient
C. Azure.AI.FormRecognizer.FormRecognizerClient
D. Azure.CognitiveServices.Speech.SpeechClient

100

Which of the following is the recommended sequence to move a QnA/Language project into a new Language service instance in a different region?

A. Export existing project → Import into new instance → Train & Publish
B. Create new instance → Copy endpoints → Delete old
C. Export → Delete → Import
D. Train first in new region then export

✅ Answer Key

B — DescribeImageInStreamAsync returns full-sentence descriptions.
A — ReadAsync then poll GetReadResultAsync loop with operation ID.
C — SpeechRecognizer is for STT.
B — Use GetCompressedFormat with MP3 container.
C — TagImageInStreamAsync returns tags.
C — loggingOptOut=true disables persistence/logging.
A — Azure.AI.Language.Conversations is correct package.
C — DocumentAnalysisClient + AnalyzeDocumentAsync("prebuilt-receipt", uri).
C — facetable enables facets/drill-down.
B — Add replicas for increased query throughput.
B — SpeechSynthesizer for TTS.
B — Polling loop with delay until status is not Running/NotStarted.
B — TranslationRecognizer is for speech translation.
C — QnAMakerDialog queries KBs.
B — log=true collects utterances for active learning.
A — DocumentAnalysisClient is the modern client.
B — FromStreamInput with compressed MP3 format.
B — AnalyzeImagesByDomainInStreamAsync for domain models.
A — azure-ai-language-conversations for Python.
C — searchable enables full-text search.
A — generateThumbnail endpoint performs smart cropping.
C — MemoryStorage is in-memory and non-persistent.
B — Avoid passing keys on CLI (they get stored in history).
C — Export for container (GZIP) is the container export.
B — StartRecognizeContentFromUri for layout/content.
B — loggingOptOut=true disables logging.
B — Read returns OperationLocation header with ID.
A — client.Features.AddPhraseListAsync(...) per snippet.
B — RecognizeOnceAsync for single utterance.
B — searchable + facetable + retrievable for those requirements.
A — /text/analytics/v3.1/languages is the language detection path.
A — FormTrainingClient used in older SDKs to train models.
B — TranslationRecognizer simplifies speech→speech translation.
B — conversationState.clear_state(turn_context) in Python snippet.
C — sortable enables sorting.
B — synthesizer.SpeakSsmlAsync or SpeakTextAsync.
C — AnalyzeImagesByDomainInStreamAsync for celebrity/landmark domain.
A — TextAnalyticsClient is the main client.
B — results.Status holds OperationStatusCodes (Running/Succeeded).
B — Configure virtual network / private endpoint settings on the resource.
A — StartRecognizeReceiptsFromUri is the receipts convenience method.
C — Orchestrator is the runtime routing component (not a dialog class).
B — /status checks container/service status and key validity.
B — Metrics Advisor offers RCA and multi-metric monitoring.
C — table projection is best for Power BI tabular export.
B — Configure language/voice, add speech resource keys, add speech to responses.
A — OperationLocation header contains the operation ID.
C — Use secure env-file or mounted secrets (not CLI args).
A — Azure.AI.Language.Conversations package for CLU.
B — StartRecognizeContentFromUri (layout/content).
A — client.Features.AddPhraseListAsync(...).
B — RecognizePiiEntitiesAsync detects PII.
B — StartContinuousRecognitionAsync for continuous realtime.
C — Descriptions provide full-sentence outputs.
A — Select latest → export for container → run/mount model.
A — QnAMakerOptions.ScoreThreshold sets min confidence.
B — Document Intelligence (Form Recognizer) for invoices.
C — Adding indexes does not reduce query throttling (replicas do).
C — loggingOptOut=true is the privacy parameter.
A — That is the documented v3.1 languages endpoint.
A — mcr.microsoft.com/azure-cognitive-services/textanalytics/sentiment.
A — StartRecognizeReceiptsFromUri(...).WaitForCompletionAsync() is the pattern.
A — _userState.CreateProperty(nameof(UserProfile));
A — Create voice in Speech Studio, then use TTS to generate audio.
C — S0 doesn’t eliminate need for Form Recognizer; statement is false.
B — CMK encryption adds metadata; index size & query time may increase.
A — Video Indexer / Video Analyzer REST APIs or SDK.
A — GetReadResultAsync(Guid) returns status/result.
A — Azure.AI.Language.Conversations package.
A — StartRecognizeContentFromUri for general layout.
B — ScoreThreshold sets the min score returned.
A — Export → Import → Train & Publish.
B — Use GetCompressedFormat for MP3 streaming audio.
A — Configure the resource's virtual network/private endpoint.
B — Describe an image from URL uses DescribeImage API; for streams there is DescribeImageInStreamAsync; older pattern is DescribeImageAsync(imageUrl). (Exam picks describe) — (B)
A — retrievable=true to be included in results.
B — MemoryStorage does NOT persist across restarts (so statement B is false).
A — QnAMakerOptions.ScoreThreshold = 0.6F.
A — TextAnalyticsClient.
A — Use an env-file with proper permissions and --env-file.
A — AnalyzeSentimentAsync performs sentiment analysis.
B — Partitions scale ingestion.
A — StartRecognizeContentFromStream (or equivalent) for content/layout.
A — POST regenerateKey with {"keyName":"Key2"} regenerates secondary key.
A — SpeechConfig.FromSubscription("key","region") is correct.
A — Set recognition language fr and add target de.
B — Attach S0 Cognitive Services for scale.
A — DocumentAnalysisClient for prebuilt invoice.
A — RecognizeOnceAsync().
A — merchantName, transactionTotal, transactionDate are typical receipt fields.
A — /swagger exposes OpenAPI docs in many containers.
A — "FO" indicates Free tier (F0) in the sample helper usage.
A — Use Video Indexer REST APIs/SDK for Azure Video Analyzer insights.
A — QnA Maker provisioning creates Cognitive Search and App Service instances.
B — StartContinuousRecognitionAsync for long audio.
A — Read secrets from a protected file/script, not pass them in CLI.
B — filterable allows OData filter expressions.
B — Include log=true in prediction requests for active learning.
A — Azure.AI.TextAnalytics.TextAnalyticsClient is correct.
A — Export → Import → Train & Publish is the standard move sequence.