Skip to content

Building a Documentation Agent with RAG

Current AI models are trained on vast datasets, making them powerful at generating general-purpose text. However, when asked about specific topics outside their training data (like your company's internal documentation), these models often hallucinate - generating plausible-sounding but incorrect information.

Thankfully, there is a solution to this problem: Retrieval-Augmented Generation (RAG). This technique consists on combining two key components:

  1. A retrieval system that finds relevant information from your custom dataset
  2. A language model that generates accurate responses using the retrieved information

In this tutorial, you'll learn how to build a RAG-powered agent that accurately answers questions about NEAR Protocol.

What You Will Need

To follow this tutorial you will need:

  1. NEAR AI CLI installed on your local machine → Installation Guide
  2. Basic understanding of creating a NEAR AI agent → Agents Quickstart Tutorial

Overview

This tutorial is divided in the following sections:

  • The problem → Understanding AI hallucination from incorrect data
  • Vector Stores → Getting started with vector stores
  • RAG Agent → Building a NEAR Documentation Q&A agent
  • Chunking → Dive deeper into how vector stores store documents
  • Embeddings → Creating document embeddings manually