Question

Your team develops a customer-support chatbot in Azure AI Foundry using a prompt flow. The lead prompt engineer wants each change to the system prompt template to go through a formal code review before it can affect the deployed flow, and wants the ability to trace which exact prompt template produced a given evaluation result. The prompt template files are currently only edited in the Foundry portal's authoring UI and never leave the workspace. What should you do FIRST to meet these requirements?

Accepted Answer

Connect the prompt flow to a Git repository and commit the flow's prompt template files, so changes are made via pull requests that require review before merging.

Answer

Enable diagnostic logging on the Foundry project and export the prompt template contents to a Log Analytics workspace after each edit.

Answer

Create a separate serverless API endpoint for each version of the prompt and switch traffic manually when a new prompt is approved.

Answer

Store the finished prompt templates in Azure Key Vault as secrets and reference them from the flow at runtime.