Skip to content

Latest commit

 

History

History
21 lines (15 loc) · 1.76 KB

TESTING_CAPABILITIES.md

File metadata and controls

21 lines (15 loc) · 1.76 KB

Testing capabilities

The Power CAT Copilot Studio Kit is a user-friendly application that empowers makers to configure agents and test sets. It has native capabilities such as Excel export or import for bulk creation and updates.

By running individual tests against the Copilot Studio APIs (Direct Line), the agent responses are evaluated against expected results. To further enrich results, additional data points can be retrieved from Azure Application Insights and from Dataverse, by analyzing Conversation Transcript records (to get the exact triggered topic name, intent recognition scores, etc.). For AI-generated answers, that are by nature non-deterministic, AI Builder prompts are used to compare the generated answer with a sample answer or with validation instructions.

Today, the tool supports these types of tests:

  • Response match
  • Attachments match
  • Topic match (requires Dataverse enrichment)
  • Generative answers (requires AI Builder for response analysis, and Azure Application Insights for details on why an answer was or was not generated)
  • Multi-turn test type is a special test type that consists of a set of test cases of regular types that are executed in specified order in same conversation context. Multi-turn tests can be used to test scenarios end-to-end, and for testing custom agents with generative orchestration.

image

image

Read more on how to configure agent for test automation

Back to the landing page