Microsoft Certified: Azure AI Engineer Associate (AZ-102)

The Microsoft Certified: Azure AI Engineer Associate certification (Exam AI-102) validates the technical skills required to design, implement, and monitor AI solutions on the Azure platform. It is an intermediate-level, role-based credential that has been recently updated to emphasize Generative AI and agentic solutions.

---------- Question 1

An Azure AI engineer is tasked with deploying a newly trained custom object detection model for identifying defects on an industrial assembly line. This model was developed using Azure AI Vision Custom Vision. The requirement is to integrate this deployment into an existing Azure DevOps continuous integration/continuous delivery CI/CD pipeline, ensure the models performance and resource consumption are continuously monitored, and manage the associated costs effectively within Azure AI Foundry. What sequence of actions and services should the AI engineer implement to meet these requirements?

1. Deploy the custom vision model manually to a container instance, then configure Azure Monitor for application performance monitoring, and set up cost alerts in the Azure portal for the container group.
2. Publish the custom vision model, integrate its deployment into the Azure DevOps CI/CD pipeline using Azure Machine Learning SDK or REST APIs, leverage Azure AI Foundry for model deployment management and monitoring via responsible AI insights, and use Azure Cost Management to track expenses.
3. Export the model to an ONNX format, create an Azure Function to host the inference, use Application Insights for monitoring, and manually review the Azure subscription bill for cost management.
4. Deploy the model directly from Custom Vision portal to an IoT Edge device, use IoT Hub metrics for performance monitoring, and estimate costs based on device usage.

---------- Question 2

A technology startup is developing an innovative AI-powered image recognition service that needs to process a high volume of user-uploaded images in real time. The service utilizes a custom-trained computer vision model. The startup anticipates fluctuating demand, with peak times requiring significant scaling and off-peak times needing to minimize costs. They also have a requirement for seamless updates to the AI model without service interruption and need to integrate this deployment into their existing DevOps pipeline. Which deployment strategy and management practices should the AI engineer recommend for their custom vision model to ensure high availability, scalability, cost-effectiveness, and smooth integration into a CI/CD pipeline within Azure AI Foundry?

1. Deploy the custom model to an Azure App Service with manual scaling and manage keys directly in application code.
2. Utilize Azure Container Instances for model deployment, implement a blue/green deployment strategy for updates, and store keys in Azure Key Vault.
3. Deploy the model as an online endpoint in Azure AI Foundry, configure auto-scaling rules based on traffic, and integrate deployment into an Azure DevOps pipeline.
4. Package the model as an Azure Function, use Consumption plan scaling, and rely on shared access signatures for authentication.

---------- Question 3

A logistics company regularly receives a wide variety of invoices and shipping manifests from numerous suppliers. These documents have diverse layouts and formats, making automated data extraction challenging. The company needs to reliably extract specific fields such as supplier name, invoice number, line items, and total amounts from all incoming documents, even those with unique structures. Which strategy should the Azure AI Engineer employ to build a robust data extraction solution that can handle the variability in document layouts?

1. Use only prebuilt models within Azure AI Document Intelligence to extract data from all documents, assuming they will cover all unique layouts.
2. Provision a Document Intelligence resource and implement a custom document intelligence model by labeling example documents for each unique layout, training, and then publishing it.
3. Create an OCR pipeline with Azure AI Content Understanding to simply extract all text from images and documents, without specific field extraction.
4. Summarize, classify, and detect attributes of documents using Azure AI Content Understanding, which is primarily focused on understanding document content rather than structured data extraction.

---------- Question 4

A telecommunications company operates a large call center and wants to improve the efficiency of their customer service representatives. They aim to implement an AI solution that can automatically transcribe customer calls in real-time, identify the intent of the caller (e.g., billing inquiry, technical support, service upgrade), and then route the call or display relevant information to the agent. The solution must support both standard and custom vocabulary specific to the telecommunications industry, ensuring high accuracy for technical terms. Which Azure AI Speech capabilities and associated custom language model components should the AI engineer utilize for this solution?

1. Text-to-speech for agent responses, and a custom speech model for accurate transcription of industry-specific terms.
2. Speech-to-text with intent recognition, using a custom language model trained with telecommunications vocabulary and utterances for specific intents.
3. Language detection for multi-language support, combined with a pre-built sentiment analysis model.
4. Speaker diarization to identify different speakers, and a custom voice model for personalized responses.

---------- Question 5

A global e-commerce company wants to enhance its customer support system by implementing an AI assistant that can both transcribe customer calls in real-time and provide natural-sounding, personalized responses via text-to-speech. The company operates in multiple countries, requiring multilingual support for transcription and translation, and frequently deals with product-specific terminology. The synthesized voice needs to maintain a consistent brand persona. What combination of Azure AI Speech capabilities should the AI engineer leverage to build this sophisticated multilingual speech-to-text and text-to-speech solution, ensuring high accuracy for specialized terminology and natural voice output?

1. Implement basic Azure AI Speech-to-Text for transcription and Azure AI Text-to-Speech with standard voices, using Azure AI Translator for multilingual support.
2. Utilize custom speech models trained on product-specific terminology for speech-to-text, implement text-to-speech using Speech Synthesis Markup Language SSML with custom neural voices, and integrate speech translation for multilingual capabilities.
3. Develop a custom language understanding model to convert spoken language to text, and use a simple audio playback system for text-to-speech responses.
4. Employ Azure AI Content Understanding to process audio streams and extract entities, then use a third-party text-to-speech API.

---------- Question 6

A large e-commerce company wants to implement a customer service chatbot that can answer complex product-related questions, assist with order tracking, and provide personalized recommendations. The chatbot needs to access a vast, constantly updated database of product information, FAQs, and customer purchase history, which resides in various structured and unstructured data stores. The primary goals are to minimize hallucinations, ensure responses are accurate and contextually relevant to the companys data, and scale to millions of users. Which generative AI approach within `Azure AI Foundry` is most appropriate to achieve these objectives?

1. Directly fine-tuning a large generative AI model like `GPT-4` on the entire product catalog and customer data, and then deploying it through `Azure OpenAI Service` to handle all customer interactions.
2. Implementing a Retrieval Augmented Generation (RAG) pattern, where `Azure AI Search` indexes the product catalog and customer data, and `Azure OpenAI` models are grounded in the retrieved information using prompt flow to generate responses.
3. Building a custom language model from scratch using `Azure Machine Learning` and training it exclusively on the company's proprietary data to ensure accuracy and relevance.
4. Utilizing only pre-trained `Azure AI Language` services for entity recognition and sentiment analysis to extract information from customer queries, and then mapping these to predefined responses without generative capabilities.

---------- Question 7

A development team is tasked with creating a novel AI-powered content summarization service using large language models within Azure AI Foundry. The solution requires rapid deployment of new models, seamless integration with existing GitHub-based continuous integration and continuous delivery CI/CD pipelines, stringent cost management, and robust security for API access keys. The team also anticipates needing to monitor model performance and resource consumption extensively after deployment. Which set of actions represents the most comprehensive and correct approach for planning, creating, deploying, managing, and securing this Azure AI Foundry service?

1. Create an Azure AI resource, select a pre-trained summarization model, deploy it using an online endpoint, integrate a custom script into GitHub Actions for CI/CD, enable cost alerts, and store account keys in Azure Key Vault.
2. Provision an Azure AI Studio project, fine-tune an existing model using Azure Machine Learning, deploy it as a batch endpoint, manually update endpoints after each deployment, set a fixed budget without monitoring, and hardcode account keys in the application.
3. Develop a proprietary summarization model, deploy it to Azure Container Instances, use Azure DevOps for CI/CD, rely on default Azure monitoring without custom dashboards, and share account keys via email.
4. Create an Azure AI Foundry resource, choose an appropriate foundation model or deploy a custom fine-tuned model as an online endpoint, integrate Azure AI Foundry into a GitHub Actions CI/CD pipeline, configure model monitoring and diagnostic settings, manage costs through Azure Cost Management tools, and utilize managed identities for authentication or secure keys with Azure Key Vault.

---------- Question 8

A large manufacturing company wants to create an intelligent automation system to optimize its production line. This system needs to monitor sensor data, identify anomalies, communicate with different machinery, order replacement parts when necessary, and alert human operators for critical issues. The solution requires complex decision-making, conditional logic, and the ability to interact with various enterprise systems autonomously. The company is exploring agentic solutions within Azure AI Foundry. Which approach or framework for building custom agents would be most suitable to handle such a sophisticated, multi-step, and autonomous manufacturing workflow?

1. Developing a simple Azure Bot Service bot with predefined command-and-response patterns for basic monitoring.
2. Utilizing Azure AI Foundry Agent Service to create and orchestrate complex agents, possibly incorporating Semantic Kernel or Autogen for advanced workflows.
3. Implementing Azure Logic Apps to manage the workflow, triggering predefined actions based on sensor data thresholds.
4. Training a single, large generative AI model to directly output all necessary actions and communications in response to raw sensor inputs.

---------- Question 9

A large university is developing an AI-powered chatbot to assist students with a wide range of inquiries, from course registration policies to campus event schedules. The chatbot needs to understand natural language student questions, provide accurate answers based on a vast and frequently updated knowledge base of university documents, and support students communicating in English, Spanish, and French. Additionally, the university wants to allow for multi-turn conversations and ensure the chatbot can handle common student slang or informal phrasing. The solution must be easily maintainable and expandable. Which set of Azure AI Language capabilities should the AI engineer employ to meet these complex requirements?

1. Create an Azure AI Language Understanding model with intents and entities, build a custom question answering project using university documents, configure it for multi-turn conversations and add alternate phrasing, then implement custom translation for multi-language support.
2. Deploy a pre-trained Azure OpenAI model for all natural language understanding and question answering, and use Azure AI Translator for all document-based responses.
3. Use Azure AI Vision to extract text from university documents, then feed it into an Azure AI Search index for basic keyword matching to answer questions.
4. Implement a simple Azure Bot Service with hardcoded rules for common questions and rely on Azure AI Speech for all language translation needs.

---------- Question 10

A telecommunications company is enhancing its call center operations by integrating AI capabilities to improve customer experience. The new system needs to perform real-time speech-to-text transcription of customer calls, translate these transcriptions into multiple languages for agents, and synthesize natural-sounding speech for automated responses, including personalized greetings. The solution must handle various accents and dialects accurately and be able to generate speech with a specific brand voice. As the Azure AI Engineer, which Azure AI Speech capabilities should you leverage to build this robust and multilingual speech processing and translation solution?

1. Use Azure AI Translator for text translation only, and integrate a third-party speech synthesis API for automated responses, ignoring custom voice requirements.
2. Implement text-to-speech and speech-to-text using Azure AI Speech, utilize Azure AI Speech for speech-to-speech and speech-to-text translation, and leverage custom voice features for brand consistency, improving accuracy with custom speech models for varied accents.
3. Manually transcribe calls and then use Azure AI Language to extract key phrases, bypassing any real-time speech processing or translation capabilities.
4. Deploy an Azure OpenAI model to generate text responses, then rely on a basic text-to-speech service without any custom voice or accent-specific improvements.
5. Implement text-to-speech and speech-to-text using Azure AI Speech, utilize Azure AI Speech for speech-to-speech and speech-to-text translation, and leverage custom voice features for brand consistency, improving accuracy with custom speech models for varied accents.

---

Are they useful?

Click here to get 360 more questions to pass this certification at the first try! Explanation for each option is included!

Follow the below LINKEDIN channel to stay updated about 89+ exams!

AI Strategy and Engineering Hub