Free Microsoft AI-102 Practice Test Questions MCQs

Explanation:
The method provisions an Azure AI service resource by kind, tier, location. FormRecognizer (Document Intelligence) is a valid kind. Standard tier is S0. eastus is a valid Azure region; useast is misspelled/invalid. The order in the dropdown is tier, location.

Correct Option Details:

FormRecognizer –
The kind specifies the service type. Form Recognizer (now Document Intelligence) extracts data from forms and invoices.

"S0", "eastus" –
S0 is the standard tier (paid). eastus is a valid Azure geography. The method expects tier then location.

Incorrect Options (why they don’t fit):

ComputerVision –
Does not specialize in document structure extraction; not optimal for most form scenarios.

CustomVision.Prediction / CustomVision.Training –
Used for custom image classification/object detection, not document analysis.

"useast", "S1" –
useast is not a valid Azure region (should be eastus, westus, etc.). S1 is not a standard tier for many services.

"S0", "useast" –
Same invalid region useast.

Reference:
Azure Cognitive Services – Resource kinds – Lists FormRecognizer as a valid kind.
Document Intelligence pricing tiers – S0 is standard tier.
Azure regions – eastus is valid; useast is not.

Explanation:
Azure Translator requires an AzureKeyCredential for authentication. The TranslationClient is the main SDK client for text translation. To translate incoming messages to English, you call the TranslateAsync method, specifying the target language (e.g., "en").

Correct Option Details:

AzureKeyCredential –
Used with TranslationClient to authenticate using the subscription key. (Other options like TranslatableObject are not real credential types.)

TranslationClient –
The correct client class for text translation in Azure Translator SDK. TranslateAgent and Translational are not valid.

client.TranslateAsync –
Asynchronous method that performs translation. client.TranslateAgent and client.Translational are not valid methods; TranslateAsync is the standard name.

Incorrect Options (why they don’t fit):

TranslatableObject / TranslateAgoric / TranstionReconqinter / Translatabale –
Typos or non‑existent classes in the Azure Translator SDK.

Client.TranslateAgent / Client.Translational / translatate.ReconqinetanceAsync –
Invalid method names. The correct method is TranslateAsync.

Reference:
Azure Translator – Text Translation client library for .NET – Shows AzureKeyCredential, TranslationClient, and TranslateAsync method.
Quickstart: Translate text with Translator – Confirms the authentication and translation call pattern.

Explanation for "Accept incoming voicemails in French":
SpeechConfig is used to configure the speech service, including the recognition language (French). It is required to set up the recognizer. However, to accept (recognize) the voicemail, you would use SpeechRecognizer with a SpeechConfig. Since SpeechRecognizer is not listed, the prerequisite SpeechConfig is the correct choice from the given options because AudioConfig handles audio input but not language settings.

Explanation for "Convert voicemails from French to English":
TranslationRecognizer is specifically designed for real-time speech translation from one language (French) to another (English). It outputs translated text and optionally synthesized speech in the target language.

Incorrect Options (why they don’t fit):

AudioConfig –
Manages audio input/output settings (e.g., microphone, file). Does not set language or perform translation.

SpeechTranslationConfig –
Used to configure translation settings (source/target languages), but the SDK class that performs the actual translation recognition is TranslationRecognizer.

SpeechSynthesizer –
Converts text to speech, not speech-to-text or translation.

TextTranslationClient –
For translating text, not speech. Requires prior speech‑to‑text conversion.

TranslationRecognitionResult –
Holds the result of translation recognition, not the class that performs the translation.

Reference:
Azure Speech SDK – TranslationRecognizer class – Real-time speech translation.
SpeechConfig for recognition language – Sets source language for recognition.

Explanation:
To identify each presenter and attribute transcribed speech to them, you need speaker indexing (also called speaker diarization). Azure AI Video Indexer can distinguish between different speakers based on voice characteristics, label each speaker with an identifier (e.g., "Speaker 1", "Speaker 2"), and provide a transcript segmented by speaker. This allows generating separate text files per presenter.

Correct Option:

C. speaker indexing
Automatically identifies and labels distinct speakers in the video.
Provides a transcript where each spoken segment is attributed to a specific speaker.
Enables generating separate text outputs per presenter.

Incorrect Options:

A. face detection –
Detects and identifies faces in video frames, but cannot segment spoken content by speaker. Faces are visual, not audio-based.

B. keyframe detection –
Identifies visually important frames (e.g., scene changes). Does not help with speaker attribution or transcription.

D. topic indexing –
Groups content by subject matter, not by speaker. Would not produce separate files per presenter.

Reference:
Azure AI Video Indexer – Speaker indexing – Explains how to identify and transcribe individual speakers.
Generate speaker‑specific transcripts – Shows JSON output with speaker IDs and attributed lines.

Explanation:
Azure AI Search indexing follows a specific pipeline order. First, document cracking extracts content from the source (e.g., PDF, blob). Then field mappings map source fields to index fields. Skillset execution runs custom skills (e.g., language & sentiment analysis). Output field mappings map skill outputs to index fields. Finally, the enriched document is pushed to the search index.

Why this order:

document cracking –
Opens and extracts text and metadata from source documents (PDFs, blobs, etc.).

field mappings –
Maps source fields (from data source) directly to index fields before any skills run.

skillset execution –
Runs the custom skill (language and sentiment analysis) on the cracked content, producing new enriched fields.

output field mappings –
Maps the skill’s output fields to the target index fields.

push to index –
Writes the final enriched document (original + skill outputs) into the search index.

Reference:
Azure AI Search enrichment pipeline order – Lists the sequence: document cracking → field mappings → skillset → output field mappings → indexing.
Custom skill integration – Skills run after field mappings but before output field mappings.

Explanation:
The requirement is to access both Speech and Language APIs using a single endpoint and credential. An Azure AI service (multi-service resource, formerly Cognitive Services multi-service account) provides a unified endpoint and key for multiple Azure AI services, including Speech, Language, Vision, and Content Safety — all under one resource.

Correct Option:

B. Azure AI service
Also known as "multi-service Cognitive Services" resource.
Provides a single endpoint (e.g., https://.cognitiveservices.azure.com/) and a single API key.
Supports Speech, Language, Vision, Decision, and Content Safety APIs from one resource.
Reduces key and endpoint management overhead.

Incorrect Options:

A. Azure AI Content Safety –
Single-service resource only for content moderation. Cannot access Speech or Language APIs.

C. Azure AI Speech –
Single-service resource only for Speech APIs (speech-to-text, text-to-speech, translation). Does not include Language APIs.

D. Azure AI Language –
Single-service resource only for Language APIs (NER, sentiment, Q&A, etc.). Does not include Speech APIs.

Reference:
Azure AI services – Multi-service resource – Confirms that a single endpoint and key provide access to multiple services including Speech and Language.
Create a multi-service resource – Step-by-step guide for provisioning.

Explanation:
The requirements — adapting to new/unforeseen challenges, improving over time based on user feedback, providing tailored recommendations — describe an autonomous agent. Autonomous agents can learn from historical data, adjust behavior patterns, and make proactive decisions without explicit prompts for every action. This matches the need for continuous improvement and adaptation.

Correct Option:

C. autonomous
Autonomous agents can incorporate user feedback to refine their recommendations and performance.
They learn from past interactions and data, enabling adaptation to new business challenges.
Provide proactive, tailored recommendations rather than just reacting to user prompts.
Designed for continuous improvement and minimal manual intervention.

Incorrect Options:

A. cognitive –
A broad, less defined term in Azure AI agent taxonomy; often overlaps with autonomous but not a standard classification. The question expects the precise type from the given list.

B. prompt-and-response –
Reactive agent that answers only when asked. Does not learn over time or adapt proactively. Cannot provide unsolicited recommendations.

D. task automation –
Executes predefined workflows (e.g., RPA). Cannot adapt to unforeseen challenges or improve from user feedback without reprogramming.

Reference:
Azure AI Agent Service – Agent types – Defines autonomous agents as capable of learning, adaptation, and proactive actions.
Autonomous agents vs. prompt-and-response – Explains that autonomous agents improve over time with feedback.

Explanation:
The requirement is to extract structured data (invoice items, sales amounts, customer details) from scanned invoice images. Azure AI Document Intelligence (formerly Form Recognizer) is specifically designed to extract key-value pairs, line items, and tables from form-like documents such as invoices, with prebuilt models requiring no custom training.

Correct Option:

B. Azure AI Document Intelligence
The prebuilt invoice model extracts invoice items (line items), sales amounts (subtotal, tax, total), and customer details (customer ID, billing address).
Works directly on scanned images and PDFs.
Minimizes development effort compared to building custom extraction.

Incorrect Options:

A. Azure AI Immersive Reader –
Designed to improve text readability for users (e.g., highlighting parts of speech, translating). Does not extract structured business data from invoices.

C. Azure AI Vision –
Performs OCR and general image analysis (tags, captions), but does not reliably identify specific invoice fields like line items or customer details without extensive post-processing.

D. Azure Custom Vision –
Used for image classification or object detection (e.g., defect detection), not for extracting text-based structured data from documents.

Reference:
Document Intelligence – Prebuilt invoice model – Shows extraction of line items, customer details, and amounts.
Choose a Document Intelligence model – Recommends the invoice model for extracting structured data from invoices.

Explanation:
The requirement involves monitoring a video feed to identify defective products using an image library of approved/rejected samples. Azure AI Custom Vision allows you to train a custom image classification or object detection model using your labeled images (defective vs. non‑defective) and then apply it to frames from a video feed. This minimizes development effort compared to building a model from scratch.

Correct Option:

B. Azure AI Custom Vision
Train a custom model using your image library of approved and rejected products.
After training, the model can analyze video frames in real time (or batch) to detect defects.
Provides a drag‑and‑drop interface for labeling and training, minimizing coding effort.
Supports classification (whole‑image pass/fail) or object detection (locating the defect).

Incorrect Options:

A. Azure AI Video Indexer –
Extracts insights like faces, labels, and transcriptions from video, but cannot train a custom defect‑detection model using your specific product images.

C. Azure Machine Learning –
More powerful but requires significantly more development effort (data preparation, model coding, deployment). Not minimal effort.

D. Azure Vision in Foundry Tools –
Refers to Computer Vision image analysis prebuilt models, which cannot be retrained with custom defect images.

Reference:
Azure Custom Vision – Overview – Designed for training custom image classification and object detection models with minimal ML expertise.
Custom Vision for manufacturing defect detection – Example use case for identifying defective products on production lines.

Explanation:
The Azure AI Agent Service allows file uploads for grounding data, code interpreter tools, and file search capabilities. According to the current service limits, the maximum total size of all files that can be uploaded per agent or per project is 100 GB. Individual file size limits also apply (typically 50 MB to 500 MB depending on the tool), but the cumulative storage quota is 100 GB.
Correct Option:

C. 100 GB

Incorrect Options:

A. 1 GB – Too low; this might be a limit for some individual file uploads, not the total across all files.

B. 10 GB – Below the documented limit for aggregate file storage.

D. 1 TB – Exceeds the current documented maximum for Azure AI Agent Service.

Reference:
Azure AI Agent Service – File upload limits – States the maximum total file storage per assistant (agent) is 100 GB.
Quotas and limits for Azure AI Agent Service – Confirms the 100 GB aggregate limit.