QuestionQ52

VerifiedEvaluating and deploying agentic workflows

You manage a support agent that uses RAG to answer questions based on a document corpus in RAG Engine. Each query retrieves the top 50 chunks. While reviewing a set of failing questions, you determine that the chunk containing the answer is among the 50 returned but is ranked near the bottom, and the agent answers from a higher-ranked chunk instead.

You want the answering chunk to rank at the top without re-indexing the corpus. The retrieval step has 200 milliseconds of headroom.

What should you do?

Explanation

The Agent Platform ranking API reranks retrieved chunks by their relevance to the query, so it can promote an answer-bearing chunk without re-indexing the corpus. Google describes the reranker as having latency of less than 100 milliseconds, making it suitable for the available 200-millisecond headroom. Reranking improves relevance but does not guarantee that the correct chunk will always rank first. (docs.cloud.google.com)

Learn more

Community Discussion

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