Skip to content

RAG Tutorial

Ever wished your AI agent could know about your private documents, company policies, or the latest updates to your knowledge base? That's exactly what RAG (Retrieval-Augmented Generation) does!

This tutorial will get you from zero to RAG-powered agent in just a few minutes. We'll cover why RAG is a game-changer, how it works behind the scenes, and most importantly—how to build your first RAG application.

Why Should You Care About RAG?

LLMs are incredibly smart, but they have some pretty big limitations:

  • No access to your private data (company docs, internal policies, recent updates)
  • Knowledge cutoff dates: They miss recent information
  • Hallucinations when they confidently make up facts

RAG solves all of these problems by giving your AI access to real, up-to-date information from your own documents.

When RAG Shines

RAG is perfect when you need:

  • Grounded answers from internal docs, policies, FAQs, or knowledge bases
  • Real-time accuracy with frequently changing information
  • Source traceability to show exactly where answers come from
  • Smart retrieval when your knowledge base is too large for prompts

When RAG Might Be Overkill

Skip RAG if:

  • You only need general world knowledge your model already has
  • Your entire knowledge base easily fits in a single prompt
  • You're doing creative writing rather than fact-based responses

How RAG Works

When you create a RAG node, RailTracks automatically handles these stages for you:

RAG Pipeline Steps

Split each document into manageable text chunks using a token-aware strategy.

Convert each chunk to a vector using an embedding model.

Write vectors and associated text/metadata to a vector store (in-memory by default).

At query time, embed the question and perform a similarity search to retrieve top-k relevant chunks.

Join retrieved snippets into a context string and pass it to an LLM for a grounded answer.

Set It and Forget It

You don't need to wire these stages yourself when using the prebuilt node—they're executed during node construction and on each query automatically!

Quickstart: Prebuilt RAG Node

This is the easiest way to add RAG to your app. Let's build a simple knowledge base in under 10 lines of code:

import asyncio
import railtracks as rt
from railtracks.prebuilt import rag_node

retriever = rt.prebuilt.rag_node([
    "Steve likes apples and enjoys them as snacks",
    "John prefers bananas for their potassium content",
    "Alice loves oranges for vitamin C",
])

question = "What does Steve like?"
results = asyncio.run(rt.call(retriever, question, top_k=3))

context = "\n".join(
    f"Document {i+1} (score: {r.score:.4f}): {r.record.text}"
    for i, r in enumerate(results)
)

print(f"Question: {question}")
print(f"Retrieved context:\n{context}")

Example Output

Your RAG node can now answer questions based on your documents:

Query: "Who is Steve?"

Response: "In our company, Steve is the lead engineer and ..."

Next Steps

Ready to build with RAG