Question

A production RAG agent deployed in Azure AI Foundry is occasionally returning incomplete answers. You need to determine, for individual failing requests, exactly which retrieval tool call returned empty context and how the model's inputs and outputs flowed through each step. You have already enabled tracing on the agent. Which action lets you pinpoint the problematic step for a specific request?

Accepted Answer

Open the individual trace in Foundry and inspect the span hierarchy, examining each span's input/output attributes and the tool-call span for the retrieval step

Answer

Review the aggregated latency and throughput dashboards to identify the average time spent in the retrieval step across all requests

Answer

Configure a groundedness evaluation on a scheduled sampling of production traffic and review the aggregate score trend

Answer

Enable Azure Monitor metric alerts on token consumption to detect when retrieval returns fewer tokens than expected