Video Transcript (English)
Welcome. In this unit, we'll learn how to use Atlas Vector Search to create a RAG or Retrieval Augmented Generation System. RAG is a well known technique used to contextualize responses that we now expect from chatbots and similar apps.
My name is Henry Weller. I'm a product manager at MongoDB, and I'll guide you through this unit. By now, we've seen that Atlas Vector Search is a powerful tool for finding semantic similarities in our data. In this unit, we'll look at a specific application of Vector Search. If you've been on the Internet in the past decade, you've likely interacted with a chatbot in some capacity.
Whether it's for customer service or one that your friends added to a group chat, chatbots have become ubiquitous.
It seems there's a chatbot for every use case. As artificial intelligence undergoes significant advancements, it's fostering the evolution of more complex and refined chatbots.
AI powered chatbots are becoming a popular tool for interacting with customers, enhancing user experience with higher efficiency and personalized responses. Even here at MongoDB, we've begun to leverage these capabilities with the release of our documentation chatbot, which answers questions about how to use MongoDB.
But how does something like the MongoDB Docs chatbot differ from ChatGPT?
Good question. The backbone of the Docs chatbot in similar applications is Retrieval Augmented Generation, or RAG for short. RAG is a natural language processing technique that enables us to provide context to the LLM so it can generate better responses. The responses use our specific data, including our documentation and MongoDB University video scripts, like the one used for this video, so they're grounded in reality. While chatbots are a common implementation of RAG, RAG systems are not only used for chatbots.
RAG is a game changer in any task where providing the right information to an LLM can make the results smarter and more precise.
We use RAG to increase accuracy no matter what the task is. Don't worry if this is new to you. We'll start from scratch and explore what RAG is by building a simple application that answers questions using content from the little book of MongoDB. To do this, we'll first learn what RAG is and how it works.
Next, we'll take a moment to learn about different AI integrations and frameworks available to help us create a RAG application. Then we'll learn how to prepare our data for the RAG application.
This involves identifying data sources, sanitizing the data, and splitting it into chunks, generating embeddings, and finally ingesting the data into MongoDB.
After that, we'll implement the retrieval component of our RAG app, which takes advantage of Atlas Vector Search. Finally, we'll implement the component that generates answers based on the question and context provided. By the end of this unit, you'll understand how RAG works and how you can leverage it in your own application. You'll also have a working RAG example that you can reference when you begin building your own GenAI applications. Are you ready to get started? I know I am. See you in the next video.