
AI-102 EXAM QUESTIONS 2026 MICROSOFT AZURE AI 102 CERTIFICATION COURSE PART-3
AI Summary
This video is part three of a series on Azure AI questions, covering important concepts for certification exams.
The first question focuses on deploying a containerized Azure Cognitive Service for sentiment analysis on an Azure Virtual Machine using Docker. The command requires specifying the container registry path for the Azure Cognitive Service Docker image. The correct path provided is `mcr.microsoft.com/azure-cognitive-services/text-analytics/sentiment`. This command utilizes backslashes for line continuation. Key parameters for running the container include CPU, memory, port exposure (5000), accepting the User License Agreement (ULA) with `accept`, providing the billing endpoint URI, and an API key for authentication.
The subsequent questions delve deeper into the Docker command parameters. For the ULA, the correct keyword to accept the terms is `accept`. For the billing parameter, the endpoint URI, such as `contoso.cognitive.services.azure.com`, must be provided. These three parameters (ULA, billing, and API key) are mandatory for the container to start.
A scenario involving an error when connecting code to Azure AI Services due to virtual network firewall rules is then addressed. The solution is to add the client's public IP address to the firewall's allow list within the Azure AI Services instance's network properties. This ensures that the specific IP address is permitted to access the service, resolving the "access denied" error.
The discussion then shifts to deploying an Azure OpenAI resource using an ARM template, specifically ensuring it can handle 600 requests per minute. The concept of "capacity" is crucial here. One capacity unit in Azure OpenAI is equivalent to 1,000 tokens per minute (TPM), which translates to 6 requests per minute (RPM). To achieve 600 RPM, 100 capacity units are required (100 capacity * 6 RPM/capacity = 600 RPM). Therefore, the ARM template property for capacity should be set to `100`.
The next question concerns regenerating the primary subscription key for an Azure AI Services resource with minimal service interruption. The recommended approach is to switch the application to use the secondary subscription key first. This allows the primary key to be regenerated without affecting ongoing service operations. Once the primary key is regenerated, the application can be switched back to use it.
Regarding visualizing metrics for Azure AI resources, the most effective way to combine charts and data from multiple resources is to add them to an Azure Dashboard. Dashboards allow for the aggregation of various visualizations and data points into a single, consolidated view.
The video then explains how to detect anomalies in sensor data from the previous 24 hours, requiring the solution to scan the entire dataset at once. This scenario calls for a "batch" detection method, as it processes a complete set of historical data rather than streaming live data or identifying individual change points.
Alerting for key regeneration events in Azure AI Services is discussed next. To create an alert that notifies when a key regeneration event occurs in the activity log, one must specify the condition with a signal type of "activity." The scope defines the resource for which the alert is set, and actions are performed after the alert is triggered.
Finally, the process of creating a composable model with Azure AI Document Intelligence is examined. Attempting to create a composable model using unlabeled custom models will result in an error. Document Intelligence requires all constituent models to be labeled when creating a composable model. This is because labels provide the necessary context for the model to understand what to predict and how to merge different models into a cohesive unit. Unlabeled data lacks this essential information, leading to the failure of the composable model creation operation.