basic-ragg

RAG (Retrieval-Augmented Generation) from scratch. 8 Jupyter notebooks building every component by hand.

Setup

# install uv (if not installed)
curl -LsSf https://astral.sh/uv/install.sh | sh

# install dependencies
uv sync

# install ollama (macOS)
brew install ollama

# pull a local LLM model
ollama pull llama3.2

# start ollama server (keep running in background)
ollama serve

Run Notebooks

uv run jupyter notebook notebooks/

Notebooks

#	Notebook	Topic	Needs Ollama?
01	01_text_and_chunking	Text splitting strategies, chunk size experiments	No
02	02_vectors_and_similarity	Dot product, cosine similarity, euclidean — from scratch	No
03	03_tfidf_from_scratch	TF-IDF vectors, keyword search, synonym problem	No
04	04_neural_embeddings	sentence-transformers, semantic search, PCA visualization	No
05	05_vector_store_from_scratch	Naive vector store, ChromaDB, scaling benchmark	No
06	06_retrieval_experiments	Chunk size impact, multi-query retrieval, score distributions	No
07	07_generation	Prompt templates, RAG vs no-RAG, hallucination test	Yes
08	08_full_rag_pipeline	Complete pipeline, interactive Q&A	Yes

Use as Module

from rag import RAGPipeline

pipe = RAGPipeline("data/knowledge_base.txt")
answer = pipe.ask("How does caffeine work?")
print(answer)

# quick test
uv run python -c "from rag import RAGPipeline; print('OK')"