Passage Retrieval (CPR) implementation overview

This is for:

Developer System Administrator

This article outlines the minimal steps required to implement Passage Retrieval (CPR) in an LLM-powered application.

Important

Passage Retrieval (CPR) is a paid product extension. Contact Coveo Sales or your Account Manager to add CPR to your organization license.

Step 1: Choose the content to use

The content that you choose for CPR will be used as the raw data from which passages are retrieved.

See Passage Retrieval (CPR) content requirements and best practices.

Note

An optimal Passage Retrieval (CPR) implementation includes both a CPR model and a Semantic Encoder (SE) model. The same SE model can be used with multiple CPR models. For best results, both the CPR and SE models should be configured to use the same content. See CPR overview for information on how CPR and SE work together in the context of a user query to retrieve passages.

Step 2: Create a CPR model

A CPR model retrieves passages based on the content specified in the model settings. The CPR model then uses the embedding vector space to find the most relevant passages (chunks) that your LLM application will use to generate the output.

See Create and manage Passage Retrieval (CPR) models.

Step 3: Associate the CPR model with a query pipeline

When a CPR model is associated with your LLM-powered application’s query pipeline, it is used to retrieve relevant passages for submitted queries.

See Associate a Passage Retrieval (CPR) model with a query pipeline.

Important

The CPR model must be associated with the same query pipeline as the Semantic Encoder (SE) model.

Step 4: Create a Semantic Encoder (SE) model

An SE model adds vector-based search capabilities to an LLM-powered application that uses CPR. Vector search allows finding items based on semantic similarity, not just keyword matches. When a query is submitted in an LLM-powered application, the top search results are sent to the CPR model. The CPR model references only these top search result items to find the most relevant passages. An SE model ensures that CPR always uses the most relevant content to retrieve passages.

See Create and manage Semantic Encoder (SE) models.

Note

An optimal Passage Retrieval (CPR) implementation includes both a CPR model and a SE model. The same SE model can be used with multiple CPR models. For best results, both the CPR and SE models should be configured to use the same content. See CPR overview for information on how CPR and SE work together in the context of a user query to retrieve passages.

Step 5: Associate the SE model with a query pipeline

When an SE model is associated with your LLM-powered application’s query pipeline, it adds vector-based search capabilities.

See Associate a Semantic Encoder (SE) model with a query pipeline.

Important

The SE model must be associated with the same query pipeline as the CPR model.

Step 6: Implement CPR

Use the Passage Retrieval API to retrieve the most relevant passages that were identified by the CPR model to enhance the generated output of your LLM application.

Note

Coveo doesn’t manage the creation of the prompt that’s sent to the LLM for output generation, or the component that displays the output.