Chapter 13: Retrieval-Augmented Generation (RAG)¶

Ground LLMs in your private data—chunking, embeddings, vector stores, hybrid search, reranking, citations, and end-to-end RAG evaluation, all running offline by default.

Metadata¶

Learning Objectives¶

• Explain why RAG: hallucination, recency, private data, context-window limits
• Implement vector similarity from scratch (cosine, top-k, in-memory index)
• Choose chunking strategies (fixed, sliding, sentence, semantic) for your data
• Use embeddings effectively and combine with TF-IDF / BM25 for hybrid search
• Apply reranking, query rewriting, HyDE, and multi-query expansion
• Evaluate RAG (hit@k, MRR, faithfulness, answer relevance) and design for production

What's Included¶

Notebooks¶

Scripts¶

• config.py — Chapter config, mock-LLM toggle, vector-store paths
• chunking.py — Fixed, sliding-window, sentence, and semantic chunkers
• vectorstore.py — InMemoryVectorStore with add, search, save, load
• rag_pipeline.py — End-to-end load → chunk → embed → retrieve → prompt → generate → cite

Exercises¶

• Problem Set 1 (notebook) — Cosine similarity from scratch, build a chunker, encode + retrieve, top-k accuracy, compare chunk sizes, source-citing prompt template
• Problem Set 2 (notebook) — BM25 + dense hybrid, query rewriting, faithfulness scorer, multi-hop retrieval, RAG evaluation harness, latency profiling
• Solutions — In exercises/solutions/ (notebooks and solutions.py for CI)

Diagrams (Mermaid)¶

• rag_architecture.mermaid, chunking_strategies.mermaid, retrieval_pipeline.mermaid

Read Online¶

• 13.1 Introduction — RAG motivation, embeddings, cosine, vector store from scratch
• 13.2 Intermediate — Chunking, embedding choices, vector stores, reranking, citations
• 13.3 Advanced — Hybrid search, query rewriting, RAG eval, production, capstone

Or try the code in the Playground.

How to Use This Chapter¶

1. Clone and install dependencies

git clone https://github.com/luigipascal/berta-chapters.git
cd berta-chapters
pip install -r requirements.txt

2. Navigate to the chapter

cd chapters/chapter-13-retrieval-augmented-generation
pip install -r requirements.txt
python -c "import nltk; nltk.download('punkt')"

3. (Optional) Install higher-quality dense embeddings and a vector DB

pip install sentence-transformers faiss-cpu chromadb

4. Launch Jupyter

jupyter notebook notebooks/01_rag_fundamentals.ipynb

Created by Luigi Pascal Rondanini | Generated by Berta AI