Plan, Create and Deploy a Microsoft Foundry Service AI-102 Practice Questions (Page 7)

Questions 31–35

1. 31expert · hard

   A robotics company is deploying a computer vision model to a fleet of autonomous robots. The robots have limited compute resources and intermittent connectivity. The model must run in real-time on the robots, and the company needs to update the model wirelessly when new versions are available. What should they do?

   • ADeploy the model as a container to Azure IoT Edge on each robot and use Azure IoT Hub for over-the-air updates.
   • BUse Azure Machine Learning to create a local web service on each robot.
   • CDeploy the model to Azure Kubernetes Service and have the robots access it over the internet.
   • DDeploy the model to Azure Container Instances and use a VPN for updates.
   Reveal Solution Discussion Select an answer first
2. 32application · medium

   A legal firm wants to use AI to summarize court documents. They need a model that can handle long documents (up to 100 pages) and provide accurate summaries. They also need to ensure the model's outputs are auditable for compliance. What should they do?

   • ADeploy a custom model to Azure Kubernetes Service to ensure performance.
   • BSelect a large language model with a long context window and enable content logging in Azure AI Foundry.
   • CSelect a small language model and split the documents into smaller chunks before summarization.
   • DUse a pre-built Azure AI Language model for summarization without additional configuration.
   Reveal Solution Discussion Select an answer first
4. 33application · medium

   A manufacturing company is deploying a predictive maintenance model using Azure AI Foundry. They want to use Azure DevOps to automate the deployment to production only after tests pass in a staging environment. They also need to roll back quickly if the model performs poorly. What should they do?

   • ADeploy directly to production from the main branch without staging.
   • BManually deploy to production after testing in staging.
   • CCreate a multi-stage pipeline in Azure DevOps that deploys to staging, runs tests, and then promotes to production with a rollback step.
   • DUse Azure Machine Learning pipelines instead of Azure DevOps because they are the only way to automate model deployment.
   Reveal Solution Discussion Select an answer first
5. 34expert · hard

   A fintech company is deploying a fraud detection model that must be updated daily with new training data. They use Azure Machine Learning for training and want to deploy the model to Azure Kubernetes Service for real-time inference. They need to automate the entire pipeline from training to deployment, and they want to ensure that a model update does not cause downtime. What should they do?

   • AUse Azure Container Instances for deployment because it supports rolling updates.
   • BCreate an Azure DevOps pipeline that trains the model, registers it in Azure Machine Learning, and deploys it to AKS using a rolling update.
   • CManually retrain the model and deploy it to AKS after hours to avoid disruption.
   • DDeploy the model to Azure Functions and use a timer trigger to update the model.
   Reveal Solution Discussion Select an answer first
6. 35expert · hard

   A news organization wants to use AI to generate article summaries. They have a requirement that the summaries must be factually accurate and not reflect any political bias. They are evaluating different models in Azure AI Foundry. What should they do?

   • AUse a fine-tuned model on the organization's own articles to ensure accuracy.
   • BSelect the largest available model to ensure the best performance.
   • CDeploy the model to Azure Kubernetes Service to improve inference speed.
   • DSelect a model with a high factual accuracy score and run bias evaluations before deployment.
   Reveal Solution Discussion Select an answer first