Table of Contents

• Overview
• 2026 RAG Landscape
• Prerequisites
• Module Map
• Module 1 — RAG Fundamentals and Architecture
• Module 2 — Document Processing and Chunking
• Module 3 — Embeddings and Vector Representations
• Module 4 — Vector Stores
• Module 5 — Basic Retrieval Techniques
• Module 6 — Advanced Retrieval Techniques
• Module 7 — Advanced RAG Patterns
• Module 8 — Agentic RAG with LangGraph
• Module 9 — Evaluating RAG with RAGAS
• Module 10 — Capstone Project with Deployment
• Getting Started
• Project Structure
• Code Examples
• Top Resources
• Contributing
• License

Overview

This repository provides a structured, step-by-step curriculum for mastering Retrieval-Augmented Generation (RAG) — one of the most impactful patterns in production AI engineering today.

RAG addresses core limitations of large language models: hallucination, knowledge cutoffs, and the inability to cite sources. This roadmap takes you from understanding those problems to building production-grade systems that solve them.

What you will build:

🌐 2026 RAG Landscape

Why RAG Is Foundational in 2026

RAG has moved from an experimental technique to the default architecture for enterprise AI systems that require accurate, traceable, and domain-specific responses. Key forces driving this:

1. Hallucination remains unsolved at scale — even frontier models hallucinate; RAG provides a grounding mechanism
2. Knowledge cutoffs still exist — retrieval gives models access to live or proprietary information
3. Regulatory requirements — industries like finance and healthcare demand source attribution and auditability
4. Context window limits — while context windows have grown, retrieval is still more cost-efficient for large corpora
5. Agentic AI requires reliable memory — RAG is the backbone of agent long-term memory and tool-augmented reasoning

RAG Architecture Evolution

Key Tools in the Ecosystem (June 2026)

🛠️ Prerequisites

Before starting this course, you should be comfortable with:

• Python fundamentals — functions, classes, file I/O, virtual environments
• GenAI fundamentals — what LLMs are, how prompting works, basic model API usage
• LangChain basics — chains, prompts, LLM wrappers (covered in the GenAI Roadmap)
• LangGraph basics — state, nodes, edges (covered in the Agentic AI Roadmap)

🗺️ Module Map

RAG Roadmap — 10 Modules
│
├── Module 1  — RAG Fundamentals and Architecture          [Week 1]
├── Module 2  — Document Processing and Chunking           [Week 2]
├── Module 3  — Embeddings and Vector Representations      [Week 3]
├── Module 4  — Vector Stores                              [Week 4]
├── Module 5  — Basic Retrieval Techniques                 [Week 5]
├── Module 6  — Advanced Retrieval Techniques              [Week 6]
├── Module 7  — Advanced RAG Patterns                      [Week 7–8]
├── Module 8  — Agentic RAG with LangGraph                 [Week 9]
├── Module 9  — Evaluating RAG with RAGAS                  [Week 10]
└── Module 10 — Capstone Project with Deployment           [Week 11–12]

📦 Module 1 — RAG Fundamentals and Architecture

Duration: ~1 week | Goal: Understand what RAG is, why it exists, and how data flows through the system

1.1 Introduction to RAG

1.1.1 What is RAG and the Problems It Solves

• Hallucination reduction — grounding generation in retrieved evidence
• Knowledge cutoff limitations — retrieving live or proprietary data the LLM was never trained on
• Domain-specific knowledge injection — plugging in enterprise or niche documents without fine-tuning
• Source attribution and transparency — every generated claim can be traced to a source chunk

1.1.2 Core Components

Knowledge Base  →  Retriever  →  Generator
     ↑                               ↓
 [Documents]               [Final Response + Sources]

1.1.3 RAG Data Flow

User Query
    │
    ▼
[Query Embedding]
    │
    ▼
[Vector Store Similarity Search]
    │
    ▼
[Top-K Relevant Chunks]
    │
    ▼
[Prompt Assembly: Query + Context]
    │
    ▼
[LLM Generation]
    │
    ▼
[Final Answer with Citations]

1.2 RAG vs Fine-Tuning vs Prompt Engineering

Decision Framework

• ✅ Use Prompt Engineering → Simple tasks, no special domain knowledge needed
• ✅ Use RAG → Dynamic, large, or proprietary knowledge bases; source attribution required
• ✅ Use Fine-Tuning → Specific output style/format; consistent domain-specific tone; fixed, curated dataset

📄 Module 2 — Document Processing and Chunking

Duration: ~1 week | Goal: Load and prepare any document type for embedding and retrieval

2.1 Document Loaders in LangChain

🧪 Hands-on: Load documents from 5 different sources (PDF, URL, CSV, JSON, plain text)

2.2 Text Splitting Strategies

RecursiveCharacterTextSplitter (Recommended Default)

• Hierarchical separator logic: ["\n\n", "\n", " ", ""]
• Parameters: chunk_size, chunk_overlap
• Respects document structure before falling back to raw character splitting

from langchain_text_splitters import RecursiveCharacterTextSplitter

splitter = RecursiveCharacterTextSplitter(
    chunk_size=512,
    chunk_overlap=50,
    separators=["\n\n", "\n", " ", ""]
)
chunks = splitter.split_documents(docs)

Other Splitters