Skip to content

Agenda — RAG Basics

Retrieval-Augmented Generation solves the most common failure mode of LLMs in production: hallucination about facts they weren't trained on. Today you build a working RAG system from scratch and understand every design decision.

Learning objectives

By the end of this session you will be able to:

  • Explain the RAG architecture and why it reduces hallucination
  • Implement five different chunking strategies and choose between them
  • Build a complete retrieval + augmentation + generation pipeline
  • Decide when RAG is better than fine-tuning (and when it isn't)

Schedule

Time Topic File
0:00 – 0:20 What is RAG? — architecture, benefits, limitations 01-what-is-rag
0:20 – 0:55 Chunking strategies — fixed, semantic, recursive, hierarchical 02-chunking-strategies
0:55 – 1:25 Retrieval and augmentation — context assembly, prompt design 03-retrieval-and-augmentation
1:25 – 2:00 RAG pipeline end-to-end — PDF ingestion to grounded answers 04-rag-pipeline-end-to-end
2:00 – 2:30 RAG vs. fine-tuning — decision framework 05-rag-vs-fine-tuning
2:30 – 3:00 Practice exercises 06-practice-exercises

Setup

pip install openai anthropic chromadb sentence-transformers pypdf2 tiktoken

import os
from openai import OpenAI
from anthropic import Anthropic
import chromadb

openai_client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
anthropic_client = Anthropic(api_key=os.getenv("ANTHROPIC_API_KEY"))
chroma_client = chromadb.Client()
print("Ready!")

← Day 02 Part 2 | Start →