AI-103: Developing AI Apps and Agents on Azure

Microsoft Foundry provides built-in response-quality evaluators (including a completeness/answer-completeness metric) that can be called from application code to assess a generated response before it is returned to the user. By adding a retry-evaluation step in the app logic—evaluating the draft response for completeness and triggering a retry (regeneration or additional retrieval) when it fails the threshold—the application enforces completeness at runtime, prior to returning the final response. This differs from adjusting model sampling parameters (temperature, presence_penalty) or swapping to a smaller model, none of which are code-level pre-return quality gates and none of which reliably guarantee response completeness.