Text reranking

Text reranking#

Use the aidb.rerank_text() function to reorder a set of candidate text passages by their relevance to a query. Reranking is typically applied as a post-retrieval step — after an initial set of candidates has been retrieved from a knowledge base — to surface the most relevant results at the top.

Step 1: Register a reranking model

SELECT aidb.create_model(
    my_reranker,
    nim_reranking,
    config => aidb.nim_reranking_config(
        api_key => nvapi-...,
        model   => nvidia/nv-rerankqa-mistral-4b-v3
    )
);

Step 2: Rerank a set of candidates

SELECT * FROM aidb.rerank_text(
    query   => What is retrieval-augmented generation?,
    inputs  => ARRAY[
        RAG combines retrieval and generation to produce grounded answers.,
        Vector databases store high-dimensional embeddings for similarity search.,
        RAG systems retrieve relevant documents before generating a response.
    ],
    options => {"model": "my_reranker"}
)
ORDER BY relevance_score DESC;

__OUTPUT__
 id | relevance_score
- ---+-----------------
  2 |        0.97
  0 |        0.91
  1 |        0.43
(3 rows)

The id column corresponds to the zero-based index of each string in the inputs array, allowing you to join back to the original data.

For the full parameter reference, see Models reference .

Using reranking after knowledge base retrieval#

A typical pattern is to retrieve an initial candidate set from a knowledge base and then rerank the results:

WITH candidates AS (
    SELECT
        source_id,
        content
    FROM aidb.retrieve_text(my_kb, What is RAG?, 20)
    JOIN my_source_table ON source_id = id::TEXT
)
SELECT
    c.source_id,
    c.content,
    r.relevance_score
FROM candidates c
JOIN aidb.rerank_text(
    query   => What is RAG?,
    inputs  => ARRAY(SELECT content FROM candidates),
    options => {"model": "my_reranker"}
) r ON r.id = (
    SELECT array_position(ARRAY(SELECT content FROM candidates), c.content) - 1
)
ORDER BY r.relevance_score DESC
LIMIT 5;

!!!Note Reranking models score passages against a query but don’t generate text. To register a reranking model, use the appropriate Model config helpers for your provider — for example, aidb.nim_reranking_config() for NVIDIA NIM reranking endpoints.

Contents