AI-102 Sample Questions & Answers
Securing and planning Azure AI Foundry services, including deployment, carries the most weight, alongside knowledge mining and information extraction, generative AI and custom agents, natural language processing, and computer vision for images and video.
Launch the full AI-102 simulator →Showing 8 of 17 free samples.
- Question 1Intermediate
You successfully run the following HTTP request.POST https://management.azure.com/subscriptions/18c51a87-3a69-47a8-aedc-a54745f708a1/resourceGroups/RG1/providers/Microsoft.CognitiveServices/accounts/contoso1/regenerateKey?api-version=2017-04-18Body{"keyName": "Key2"}What is the result of the request?
Show answer & explanation
Correct answer: B
- Question 2Intermediate
Implement an agentic solution · Implement function calling
You are designing an autonomous customer support solution for a telecommunications company using Azure AI Agent Service. The system must autonomously handle complex user refunds by looking up customer data, calculating refund eligibility based on usage logs, and submitting refund requests to a legacy payment system via REST API. The agent must be able to reason through the steps without a hardcoded workflow. Which component should you implement to enable the agent to interact with the external payment system?
Show answer & explanation
Correct answer: A
To enable an AI agent to interact with external systems, you must define tools (often using function calling or plugin definitions). Providing an OpenAPI specification allows the agent to understand the API structure, parameters, and return values, enabling it to construct the correct API calls dynamically based on its reasoning.
- Question 3Advanced
Implement an agentic solution · Design agentic AI solutions
You are implementing a multi-agent system using the AutoGen framework to generate software code. You need to configure a specific agent that executes the code generated by the 'Assistant' agent to validate it. This agent should run within a Docker container to prevent malicious code execution on the host system. Which type of agent configuration should you use?
Show answer & explanation
Correct answer: C
In AutoGen, the UserProxyAgent is typically responsible for executing code. To ensure security and isolation, you configure the
code_execution_configparameter to specify a Docker container image. This ensures that any code generated by the Assistant agent runs in an isolated environment. - Question 4Intermediate
Implement generative AI solutions · Implement retrieval-augmented generation (RAG)
You are optimizing a Retrieval-Augmented Generation (RAG) solution in Azure AI Foundry. Users report that the model answers are accurate but often lack cohesion when summarizing long documents split into many small chunks. You need to improve the flow of the answers without losing the detail provided by the chunks. What should you adjust in your chunking strategy?
Show answer & explanation
Correct answer: B
Increasing the overlap between chunks ensures that context is preserved across chunk boundaries. This helps the LLM generate more cohesive answers because adjacent chunks share bridging information, reducing disjointed transitions in the retrieved context.
- Question 5Intermediate
Implement knowledge mining and information extraction solutions · Implement an Azure Cognitive Search solution
You are configuring an Azure AI Search index to support a RAG application. You want to ensure that the search results prioritize documents that contain the exact keywords from the user query, but also include conceptually related documents even if the keywords don't match exactly. Which search configuration should you implement?
Show answer & explanation
Correct answer: D
Hybrid search combines the precision of keyword search (BM25) with the conceptual understanding of vector search. Reciprocal Rank Fusion (RRF) is the algorithm used to merge the ranked results from both streams into a single unified result set, satisfying both requirements.
- Question 6Intermediate
Implement generative AI solutions · Implement solutions by using Azure OpenAI Service
A developer is using the Azure AI Foundry SDK to evaluate a chat flow. They want to measure how well the generated answer aligns with the retrieved context data to ensure the model is not hallucinating information. Which evaluation metric should they select?
Show answer & explanation
Correct answer: A
Groundedness measures the extent to which the model's response is supported by the provided source context. A high groundedness score indicates the model is sticking to the retrieved data and not hallucinating external or false information.
- Question 7Advanced
Plan and manage an Azure AI solution · Manage costs for Azure AI services
You are managing a high-traffic Azure OpenAI resource. You need to ensure that your application maintains consistent performance and is not subject to the noisy neighbor problem or fluctuating latency during peak hours. You also need a predictable monthly cost. Which deployment model should you choose?
Show answer & explanation
Correct answer: A
Provisioned Throughput Units (PTU) provide reserved capacity for Azure OpenAI models. This ensures consistent latency and throughput (avoiding noisy neighbors) and offers a predictable pricing model compared to the token-based Pay-As-You-Go model.
- Question 8Advanced
Implement knowledge mining and information extraction solutions · Enrich documents using cognitive skills
You are implementing a custom skill in Azure AI Search that calls an Azure Function. The function processes complex documents and can take up to 3 minutes to complete. The indexer is failing with a timeout error. Which action should you take to resolve this?
Show answer & explanation
Correct answer: C
The default timeout for a custom Web API skill is 30 seconds. To support longer running functions, you must explicitly set the
timeoutproperty in the skill definition (e.g., in ISO 8601 format like PT3M30S) to allow the indexer to wait longer for the response.
Ready for the real thing?
The full AI-102 simulator has every exam-style question, timed mode, and instant scoring.