Integrate Agents with Azure AB-620 Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A manufacturing company is creating a Copilot Studio agent to assist engineers with troubleshooting equipment manuals. The manuals are stored as PDFs in a SharePoint document library. The team wants to use Azure AI Search to provide generative answers. They have already created an Azure AI Search service and an index. What should they do next to ensure the PDFs are searchable?

   • AEnable Application Insights to monitor the search index and automatically ingest new PDFs.
   • BUse the Foundry model catalog to create a custom skill that reads the PDFs and writes to the index.
   • CUpload the PDFs directly to the Azure AI Search index using the Azure portal.
   • DCreate an indexer that connects to SharePoint and runs on a schedule to ingest the PDFs.
   Reveal Solution Discussion Select an answer first
2. 12expert · hard

   A consulting firm is using a Copilot Studio agent that uses a custom prompt from Foundry to generate client reports. The prompt includes confidential client information that is retrieved from Azure AI Search. The firm wants to use Application Insights to monitor the agent, but they are concerned about the confidentiality of the data in the logs. What should they do?

   • AUse Azure AI Search logs instead of Application Insights.
   • BLog the full prompt and response in Application Insights and restrict access to the logs.
   • CDisable Application Insights and use manual monitoring.
   • DUse custom properties in Application Insights to log only non-sensitive metadata, such as the prompt version and search result count, and avoid logging the actual content.
   Reveal Solution Discussion Select an answer first
4. 13application · medium

   A retail company is deploying a Copilot Studio agent that helps customers track orders. They want to use the Foundry model catalog to configure a custom prompt that includes the user's order ID and the order status from an Azure SQL database. They also want to monitor the agent's performance using Application Insights. The team needs to ensure that the prompt is grounded with the order data. What should they do?

   • AUse the Foundry model catalog to create a prompt that calls an Azure Function to fetch order data, and then use that data in the prompt.
   • BEnable Application Insights and use its query editor to fetch order data for the prompt.
   • CConfigure Azure AI Search with the order data and use that as the only knowledge source for the prompt.
   • DAdd a custom connector to the agent that calls the Azure SQL database, and then reference the connector's output in the Foundry prompt.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A large enterprise is deploying a Copilot Studio agent that uses Azure AI Search to answer employee questions from a mix of structured HR data (in SQL) and unstructured policy documents (in PDFs). The team wants to use the Foundry model catalog to create a custom prompt that can answer questions like 'How many vacation days do I have?' and 'What is the remote work policy?'. They need to ensure that the agent provides accurate answers and can handle both types of queries. What is the best approach?

   • AUse the Foundry model catalog to create a prompt that directly queries the SQL database and ignores Azure AI Search.
   • BUse Azure AI Search to index both the SQL data and the PDFs, and use the Foundry prompt to instruct the model to use the search results for both types of queries.
   • CUse a custom connector to query the SQL database for structured queries and use Azure AI Search for unstructured documents, with the Foundry prompt deciding which source to use.
   • DUse Azure AI Search to index the SQL data and the PDFs, and enable Application Insights to monitor the accuracy of responses.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   An IT service desk is using a Copilot Studio agent that leverages a custom prompt from the Foundry model catalog to generate troubleshooting steps. The team notices that the agent sometimes gives incorrect steps. They want to use Application Insights to identify which prompts are associated with incorrect answers. What should they do?

   • ALog only the user's query in Application Insights.
   • BLog only the agent's response in Application Insights.
   • CLog the prompt version and the user's follow-up correction or rating in Application Insights.
   • DUse the Foundry model catalog to review the prompt's training data.
   Reveal Solution Discussion Select an answer first