Artificial Intelligence (AI) · Intermediate level

Intermediate level · Module 4: Retrieval-Augmented Generation · Lesson 3 of 3

Evaluating Retrieval and Grounding

Measure whether the right passages are found and whether answers stay faithful to them.

In this lesson you will

  • Measure retrieval hit rate
  • Check answers for grounding and correct citations

Key concepts

retrieval hit rategroundingcitation accuracyunanswerable questions

The lesson

Evaluate retrieval separately: for each test question, is the passage containing the answer among the retrieved chunks? This hit rate shows whether failures come from search or from generation.

Check grounding: every claim in the answer should be supported by a cited passage. Unsupported claims are hallucinations even if they happen to be true.

Test unanswerable questions. A good assistant says it does not know and points to a human contact instead of inventing an answer.

Step by step

  1. Label the correct passage for each test question.
  2. Measure the retrieval hit rate.
  3. Check answers for grounding and citations.
  4. Test five unanswerable questions.

Practice activity

Evaluate and improve the FAQ assistant's retrieval and grounding.

Expected result: Measured hit rate and grounding results, with at least one improvement made.

Check your understanding

  1. Why evaluate retrieval separately from answers?

    Show answer

    To find out whether failures come from search or generation.

  2. What should the assistant do with an unanswerable question?

    Show answer

    Say it does not know and point to a human contact.

Taught live by an instructor — online or in person. Free course lessons are for self-study and do not lead to a certificate — see registration options.