// the find
athina-ai/rag-cookbooks
This repository contains various advanced techniques for Retrieval-Augmented Generation (RAG) systems.
A collection of Jupyter notebooks implementing RAG patterns from naive retrieval up through agentic variants (Self-RAG, Corrective RAG, Adaptive RAG). It's aimed at developers who want working code to study or adapt rather than reading papers cold. Athina AI's evaluation library is woven into every notebook, which is both useful and a recurring sales pitch.
Each notebook is self-contained and runnable in Colab with no local setup, which lowers the barrier significantly for experimentation. The progression from naive → hybrid → agentic is logical and the research paper references alongside each technique are genuinely helpful for understanding why a technique exists, not just how to implement it. Coverage of less-obvious patterns like HyDE (hypothetical document embeddings) and RAG Fusion with Reciprocal Rank Fusion is more useful than yet another basic retrieval demo. The unstructured RAG notebook handling mixed text/tables/images is one of the harder problems and it's good to see it included.
Every single notebook is coupled to Athina AI for evaluation, which means you're learning their SDK as much as you're learning RAG — there's no path to evaluating without a vendor account. The repo hasn't been touched since February 2025 and the LangChain API surface changes constantly, so some notebooks are likely already broken. There are no shared utilities across notebooks; chunking, embedding, and retrieval setup is copy-pasted everywhere, making it hard to compare techniques fairly since the baseline conditions differ. The dataset used throughout is a Tesla Q3 earnings PDF, which is fine for a demo but doesn't help you understand how these techniques behave on code, structured data, or multi-language content.