Struct PostgresVectorRetriever

pub struct PostgresVectorRetriever<T>
where
    T: AsyncEmbeddingClient,
{ /* private fields */ }

PostgresVectorRetriever

This struct is a allows for the retrieval of similar text from a postgres database. It is parameterized over a type T which implements the AsyncEmbeddingClient trait. This is because text needs to be embeded before it can be compared to other text. You must connect first create a PostgresVectorStore as this handles connecting to the database. then you can calle .as_retriever() to convert it to retriever.

Examples

use rag_toolchain::retrievers::*;
use rag_toolchain::clients::*;
use rag_toolchain::common::*;
use rag_toolchain::stores::*;
use std::num::NonZeroU32;

async fn retrieve() {
    let chunk: Chunk = Chunk::new("This is the text you want to retrieve something similar to");
    let top_k: NonZeroU32 = NonZeroU32::new(5).unwrap();
    let distance_function: DistanceFunction = DistanceFunction::Cosine;
    let embedding_model: OpenAIEmbeddingModel = OpenAIEmbeddingModel::TextEmbedding3Small;
    let client: OpenAIEmbeddingClient = OpenAIEmbeddingClient::try_new(embedding_model).unwrap();
    let store: PostgresVectorStore = PostgresVectorStore::try_new("table_name", embedding_model).await.unwrap();
    let retriever: PostgresVectorRetriever<OpenAIEmbeddingClient> = store.as_retriever(client, distance_function);
    // This will return the top 5 most similar chunks to the input text.
    let similar_text: Chunks = retriever.retrieve(chunk.content(), top_k).await.unwrap();
}

Trait Implementations

impl<T> AsyncRetriever for PostgresVectorRetriever<T>

where T: AsyncEmbeddingClient + Sync, T::ErrorType: 'static,

async fn retrieve( &self, text: &str, top_k: NonZeroU32, ) -> Result<Chunks, Self::ErrorType>

PostgresVectorRetriever::retrieve

Implementation of the retrieve function for PostgresVectorRetriever. This allows us to retrieve similar text from the vector database.

Arguments

• text: &str - The text we are searching for similar text against.
• top_k: NonZeroU32 - The number of results to return.

Errors

• PostgresRetrieverError::EmbeddingClientError - If the embedding client returns an error.
• PostgresRetrieverError::QueryError - If there is an error querying the database.

Returns

• Chunks which are the most similar to the input text.

type ErrorType = PostgresRetrieverError<<T as AsyncEmbeddingClient>::ErrorType>

Custom error type for the retriever