Construye un asistente de conocimiento RAG

Learn how to build a Retrieval-Augmented Generation (RAG) assistant that answers questions using your own documents. In this guide, you'll index content with embeddings, perform semantic search, and generate accurate, context-aware responses.

Overview

A RAG (Retrieval-Augmented Generation) application combines a language model with a knowledge base. Instead of relying only on the model's training data, relevant documents are retrieved first and then provided as context for generating an answer.

This approach improves accuracy and enables AI applications to answer questions using your own data.

Estimated time: 20–30 minutes

Prerequisites

Before starting, ensure you have:

• A NovaAI project

• API authentication configured

• Documents to index

• Access to the Embeddings and Vector Store APIs

RAG Architecture

Supported examples include:

• Product documentation

• Knowledge bases

• User manuals

• Internal company documents

• FAQs

1const embedding = await client.embeddings.create({
2  model: "nova-embed-1",
3  input: documentText
4});

1const embedding = await client.embeddings.create({
2  model: "nova-embed-1",
3  input: documentText
4});

Knowledge Base
├── Product Docs
├── API Reference
├── FAQs
├── Tutorials
└── Release Notes

Knowledge Base
├── Product Docs
├── API Reference
├── FAQs
├── Tutorials
└── Release Notes

1const results = await client.vectorStore.search({
2  query: userQuestion,
3  top_k: 5
4});

1const results = await client.vectorStore.search({
2  query: userQuestion,
3  top_k: 5
4});

1const response = await client.chat.create({
2  model: "nova-chat-4",
3  messages: [
4    {
5      role: "system",
6      content: "Answer only using the provided documentation."
7    },
8    {
9      role: "user",
10      content: userQuestion
11    }
12  ]
13});

1const response = await client.chat.create({
2  model: "nova-chat-4",
3  messages: [
4    {
5      role: "system",
6      content: "Answer only using the provided documentation."
7    },
8    {
9      role: "user",
10      content: userQuestion
11    }
12  ]
13});

Best Practices

Completed

Congratulations!

You've built a Retrieval-Augmented Generation assistant capable of answering questions using your own knowledge base.