Module 19: Retrieval-Augmented Generation (RAG)

Chapter Overview

Large language models are powerful generators but inherently limited by their training data cutoff, their tendency to hallucinate, and the impossibility of encoding all world knowledge in model parameters. Retrieval-Augmented Generation (RAG) addresses these limitations by connecting LLMs to external knowledge sources at inference time, grounding responses in retrieved evidence rather than relying solely on parametric memory.

This module covers the complete RAG landscape, from fundamental architectures through advanced retrieval techniques. You will learn how to build ingestion pipelines, implement query transformations, combine dense and sparse retrieval, and leverage knowledge graphs for structured reasoning. The module also explores agentic RAG systems that can decompose complex queries, perform iterative research, and synthesize information from multiple sources.

On the structured data side, you will learn how LLMs can query databases through text-to-SQL, process tabular data, and combine structured and unstructured retrieval. Finally, the module surveys the major RAG frameworks (LangChain, LlamaIndex, Haystack) that provide production-ready tooling for building retrieval-augmented applications.

Learning Objectives

Design and implement end-to-end RAG pipelines including document ingestion, chunking, embedding, and retrieval
Apply advanced retrieval techniques such as HyDE, multi-query expansion, cross-encoder re-ranking, and fusion retrieval
Construct and query knowledge graphs for structured reasoning, including GraphRAG with community detection
Build agentic RAG systems capable of query decomposition, iterative research, and multi-source synthesis
Implement text-to-SQL pipelines for structured data retrieval with schema linking and error correction
Evaluate RAG system quality using faithfulness, relevance, and answer correctness metrics
Compare and use RAG orchestration frameworks (LangChain, LlamaIndex, Haystack) for production applications
Diagnose and fix common RAG failure modes including lost-in-the-middle effects, retrieval drift, and context window overflow

Prerequisites

Module 18: Embeddings & Vector Databases (embedding models, similarity search, vector stores)
Module 09: LLM APIs (calling OpenAI, Anthropic, and other providers programmatically)
Module 10: Prompt Engineering (system prompts, few-shot examples, structured outputs)
Familiarity with Python, including working with APIs and JSON data
Basic understanding of SQL and relational databases (for Section 19.5)

Chapter Overview

Learning Objectives

Prerequisites

Sections