QuestionQ39

VerifiedEvaluating and deploying agentic workflows

You are preparing to deploy an enterprise technical troubleshooting agent built with Agent Development Kit (ADK). Your company’s QA team has compiled 500 representative customer support test cases paired with authoritative reference documents and expected answer trajectories. You need to establish a continuous integration testing process to measure document retrieval precision and answer faithfulness before releasing software updates.

How should you configure the evaluation pipelines in ADK using the agents-cli eval run command?

Explanation

Context precision and recall compare retrieved documents with ground-truth documents, measuring retrieval relevance and coverage. They do not establish whether the answer is faithful to those documents, so the pipeline also needs a separate grounding or hallucination metric. The agents-cli evaluation guide recommends such metrics for retrieval-augmented agents. (google.github.io)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!