AI/ML
RAG From Scratch
A 13-module RAG system covering document ingestion, chunking, vector search, reranking, and conversational RAG.
Tech Stack
The Problem
Most RAG tutorials show one happy-path pipeline and skip the decisions that actually matter: how to chunk documents, which retrieval strategy to use, and how the pieces fit together.
How It Works
Thirteen progressive modules in Python notebooks. The first four build the core pipeline from document ingestion to conversational RAG with chat history. The middle modules compare chunking strategies: character-based, recursive, semantic, and LLM-driven. The later modules cover multi-query retrieval, reciprocal rank fusion, hybrid search, and reranking with Cohere. Everything runs locally with Llama 3 through Ollama and ChromaDB for vector storage.