Adding Retrieval-Augmented Generation to the chat

In this lesson we'll deepen our understanding of how to make our chat applications smarter by integrating retrieval-augmented generation. 

Previously, we set up a basic chatbot using a FastAPI, LangChain and Server-Sent Events. We also built a semantic index using the IPCC report. This time, we're going to use that indexed information dynamically during our chat sessions.

We use our previous implementation as as starting point. When our chat receives a query that may need external information (like data from the IPCC report), the system performs the following steps:

---
title: Adding Retrieval-Augmented Generation to the chat
isPublicLesson: true
privateVideoUrl: https://stream.mux.com/Aws6SJqsL02rwwOX6P02X2OIAjS1BhwuRlKUfaqmbVvPw.m3u8
---

# Adding Retrieval-Augmented Generation to the chat

In this lesson we'll deepen our understanding of how to make our chat applications smarter by integrating retrieval-augmented generation. 

Previously, we set up a basic chatbot using a FastAPI, LangChain and Server-Sent Events. We also built a semantic index using the IPCC report. This time, we're going to use that indexed information dynamically during our chat sessions.

We use our previous implementation as as starting point. When our chat receives a query that may need external information (like data from the IPCC report), the system performs the following steps:

1.  **Initial Query to Model**:

    -   We use our `ChatOpenAI` instance to determine if the user's question relates to the IPCC report.
    -   If so, the model is prompted to construct a query to retrieve relevant information.
2.  **Fetching Data**:

    -   Our system fetches the relevant data using the query provided by the model.
    -   This step involves making a semantic search over the index.
3.  **Integration into Chat**:

    -   The retrieved data is then formatted (using YAML for better comprehension by the ChatGPT model) and added to the ongoing chat context as a `function` message.
    -   This allows the model to use this context to generate more informed and accurate responses.

Louis Sanna

Welcome back. In this lesson, we will use the retrieval augmented generation

In the previous lesson, we build a semantic index using the IPCC report.

Now, we are going to use the right river doing regeneration.

So, we are starting from the previous chat implementation.

So, exactly as time, we have a Python class, we have a fast App endpoint,

and we are using the server set event once again.

All this code is exactly as the same as was explained in the backend module.

But now, before we do our generation, we need to fetch some content.

This is our new code. Now, how does it work?

We need to check whether the red river is needed to answer the user question.

How do we do that? We make another code to the realm, and we ask,

Does the user need to run version from the IPCC report?

So, we get the query, we once again use the right river using await,

And now, we can pass to the model the extract using a message with the whole

So, we set whole function, we set the name of the method,

Note that we use the ML because charge Bt is slightly better at understanding

And now, if we do the same generation, the model will be able to use the chunk

to make it on to more precise and more reliable.

Note that you can also emit the event support you would like to display the

You can emit the event exactly as we did for an LMS chunk,

and it will be sent to the front end and you just need to build the UI as you

We finished implementing the retrieval augmented generation,

but turn on the congratulations and see you soon.

Learn

The newline Guide to Building Your First GraphQL Server with Node and TypeScript

Teach

Amelia Wattenberger

Author of Fullstack D3

Community

Masterclasses

Tutorials

Fullstack React with TypeScript