Answer questions from your policies
A customer asks how to get a refund. Search the support collection, retrieve the relevant policy, and give a language model those passages to draft an answer a support agent can check.

Find the passage behind the answer
Start with five documents, filter to support policies published since January 2026, then rank the remaining rows by their embedding distance from the question. Filtering keeps an unrelated financial or legal document out of the answer context.
| Document | Cosine distance |
|---|---|
| Refund Policy | 0.5137 |
| Shipping FAQ | 0.8524 |
The smaller distance puts the refund policy first. The agent can check the wording of the proposed answer against that passage. Run the vector search and retrieval tutorial
Combine keywords with meaning
For a technical knowledge base, exact terms can matter as much as meaning. Rank articles once with BM25 and once by embedding distance. Reciprocal rank fusion combines their positions into one list.
