Identify Fundamental AI Concepts and Model Components
Identify Common AI Workloads and Capabilities
Build Generative AI Apps and Agents with Microsoft Foundry
Implement Text and Speech Solutions with Microsoft Foundry
Implement Vision and Image-Generation Solutions with Microsoft Foundry
Extract Information from Content with Microsoft Foundry
Sample questions with answers
8 of the 38 questions in this set, with the correct answer marked and every option explained.
1. You are designing an application that uses Azure OpenAI Service to analyze customer support chats and extract key entities like names, dates, and product issues. Which deployment model should you use to ensure your data is not used to train Microsoft's foundational models?
✗You should use Azure Cognitive Search instead.
A different service entirely, for indexing and retrieval. It does not address how Azure OpenAI handles your prompts.
✗You must explicitly opt out via an Azure support ticket.
No opt-out is needed, because the behaviour is the default. Microsoft states that customer data submitted to Azure OpenAI is not used to train the foundational models.
✗You must use a dedicated Azure OpenAI cluster.
Provisioned throughput buys you reserved capacity and predictable latency. It is a performance and cost option, not a data-privacy control — the privacy commitment applies to standard deployments too.
✓All Azure OpenAI Service deployments ensure customer data is not used to train foundational models by default.
By default, Microsoft does not use customer data (prompts, completions, or training data) submitted to Azure OpenAI Service to train its foundational models.
2. You need to implement a Retrieval-Augmented Generation (RAG) pattern to ground an Azure OpenAI model on your company's proprietary PDF documents. Which Azure service is the recommended vector database for storing the document embeddings?
✓Azure AI Search (formerly Cognitive Search)
Azure AI Search (formerly Azure Cognitive Search) provides enterprise-grade vector search capabilities and integrates seamlessly with Azure OpenAI Service to implement the RAG pattern.
✗Azure SQL Database
A relational database. It can store embeddings as data, but it is not the recommended vector store for a RAG pipeline and lacks integrated document cracking and chunking.
✗Azure Blob Storage
Where the source PDFs would live. Object storage holds files; it performs no vector similarity search.
✗Azure Cosmos DB (MongoDB vCore or NoSQL)
Genuinely supports vector search and is a legitimate choice, particularly if your application data already lives there. Azure AI Search is the recommended default for document RAG because it bundles ingestion, chunking, hybrid keyword-plus-vector retrieval and semantic ranking.
3. Your application uses Azure OpenAI Service. You notice that the model sometimes generates harmful or inappropriate content. You want to implement a solution to automatically filter out hate speech and self-harm content from both user inputs and model outputs. What should you configure?
✗Azure Active Directory Conditional Access
Controls who may sign in and under what conditions. Identity and access, with no visibility into content.
✗Azure Web Application Firewall (WAF)
Blocks malicious HTTP traffic such as SQL injection and cross-site scripting at the network edge. It inspects request structure, not the meaning of generated text.
✓Azure AI Content Safety
Azure AI Content Safety detects harmful user-generated and AI-generated content in applications, providing severity scores across categories like hate, sexual, violence, and self-harm.
✗Azure Monitor Log Analytics
Collects and queries telemetry. It could record that harmful content occurred but cannot prevent it from being returned.
4. You are building a custom copilot application using Semantic Kernel. What is the primary purpose of a 'plugin' within the Semantic Kernel framework?
✓To provide the AI with native functions (like database lookups or API calls) it can use to perform actions.
In Semantic Kernel, plugins encapsulate capabilities (like native C# or Python functions, or other AI prompts) that the kernel can orchestrate to accomplish tasks.
✗To compress the prompt before sending it to Azure OpenAI.
Prompt size is managed through chunking and context strategies. Plugins extend capability rather than reduce token count.
✗To host the LLM weights locally.
Semantic Kernel is an orchestration SDK. It calls a model hosted elsewhere and never holds weights.
✗To monitor billing and token usage.
Cost and consumption are tracked through Azure Monitor and the portal. Plugins are about what the AI can do, not what it costs.
5. You are using the Azure OpenAI API to generate Python code. You want the model to return the absolute most likely sequence of tokens without any randomness. Which parameter should you adjust?
✗Set max_tokens to 100
Caps the length of the response. A shorter answer is no more deterministic than a long one.
✓Set temperature to 0
Setting the temperature to 0 makes the model highly deterministic, always picking the most probable next token. This is ideal for tasks requiring strict formatting, like code generation.
✗Set presence_penalty to 1
Discourages reusing tokens that have already appeared, pushing the model toward new topics. It changes what is generated but adds variability rather than removing it.
✗Set frequency_penalty to 1
Penalises tokens in proportion to how often they have already occurred, reducing repetition. Again a content control, not a determinism control.
6. You need to extract printed text, handwriting, and structural elements (like tables and checkboxes) from scanned financial forms. Which Azure AI service is specifically designed for this?
✗Azure AI Vision (OCR)
Reads printed and handwritten text from images, so it covers part of the requirement. What it does not do is recover structure — tables, checkboxes and key-value pairs are Document Intelligence's contribution.
✗Azure AI Language
Analyses text that has already been extracted, for sentiment, entities and summarisation. It cannot read a scanned page.
✓Azure AI Document Intelligence (formerly Form Recognizer)
Azure AI Document Intelligence uses advanced machine learning to extract text, key-value pairs, tables, and structures from documents. It goes beyond simple OCR by understanding document layout.
✗Azure AI Custom Vision
Trains image classification and object detection models on your own labelled images. It recognises objects, not document text and layout.
7. Your company receives thousands of customer reviews daily. You need to build a system that automatically categorizes these reviews into 'Pricing', 'Quality', or 'Customer Service', and also determines if the tone is positive or negative. Which service should you use?
✗Azure AI Translator
Converts text between languages. It preserves meaning across languages without classifying or scoring it.
✗Azure AI Speech
Handles speech-to-text and text-to-speech. The reviews here are already text.
✗Azure OpenAI Service exclusively
Could certainly do both tasks through prompting, which is why 'exclusively' is the problem. For well-defined classification and sentiment at high volume, Azure AI Language offers purpose-built models at lower cost and more predictable latency.
✓Azure AI Language
Azure AI Language provides pre-built and custom capabilities for Text Analytics, including sentiment analysis and custom text classification, making it the most direct and cost-effective service for this.
8. Microsoft outlines six guiding principles for Responsible AI. Which principle ensures that AI systems treat all people fairly and avoid allocating resources or opportunities differently based on demographic attributes?
✗Inclusiveness
Concerns designing so that people of all abilities and backgrounds can use the system — accessibility and participation. Related to fairness but focused on empowerment rather than equitable allocation.
✗Privacy and security
Covers protecting personal data and securing the system. A different obligation from equitable treatment.
✗Reliability and safety
Requires that systems perform consistently and fail safely. A system can be entirely reliable and still allocate outcomes unfairly.
✓Fairness
The Fairness principle dictates that AI systems should treat everyone fairly and avoid affecting similarly situated groups of people in different ways.
30 more questions in the app
Practise the full 38-question set with a timer, scoring and progress tracking.
