Evaluation

Check Evaluation results in the Portal 

Use the Azure AI Foundry portal to inspect individual evaluation runs:

• Go to the Evaluation page.
• Locate your evaluation run in the run list.
• Click on the run to open the run detail page.
• Review each data sample and its associated metrics, including any null or failed entries

View Evaluation Results in the Azure AI Foundry Portal - Azure AI Foundry | Microsoft Learn

Check Evaluation Configuration and Data Mapping

Groundedness evaluation requires specific fields:

• Query: Required
• Response: Required
• Context: Required

If any of these are missing or malformed in your dataset, the evaluator may return null.

Evaluate Generative AI Models and Apps with Azure AI Foundry - Azure AI Foundry | Microsoft Learn

Inspect Dataset Format

Ensure your dataset:

• Is in JSONL or CSV format
• Contains valid entries for query, response, and context
• Avoids empty strings or missing fields

You can preview your dataset during evaluation setup in the portal to catch formatting issues early

Use the SDK for Local Degugging

If portal logs aren’t sufficient, use the Azure AI Evaluation SDK to run evaluations locally:

Python:  pip install azure-ai-evaluation

This lets you isolate problematic entries and inspect detailed error messages.

Local Evaluation with the Azure AI Evaluation SDK - Azure AI Foundry | Microsoft Learn

Monitor Application Health

If the issue is related to runtime behavior or model deployment, use the Monitoring tab in Azure AI Foundry:

• Navigate to Monitoring > Application Analytics
• Use filters to inspect logs and metrics
• If needed, open the view in Azure Monitor Application Insights for deeper analysis

Monitor your Generative AI Applications - Azure AI Foundry | Microsoft Learn