// the find
bragai/bRAG-langchain
Everything you need to know to build your own RAG application
A collection of five Jupyter notebooks walking through RAG patterns from basic vector search to multi-query, routing, multi-representation indexing, and reranking. It's aimed at developers who learn best by running code rather than reading docs, and who want to go from zero to a working LangChain RAG pipeline in a day or two. Explicitly a tutorial, not a library.
The progression is well-structured — each notebook builds on the last without assuming you ran the previous one perfectly. Notebook 3's coverage of both logical and semantic routing is genuinely useful and not something most intro RAG tutorials bother with. Notebook 5 includes Reciprocal Rank Fusion and Cohere reranking side-by-side, which gives you a practical comparison rather than just theory. The boilerplate `full_basic_rag.ipynb` is a good starting point you can actually copy-paste into a real project.
Everything is hardwired to OpenAI embeddings and LangChain abstractions — if you want to swap in a different embedding model or drop LangChain, you're mostly on your own. The notebooks require four separate paid API keys (OpenAI, Pinecone, Cohere, LangSmith) before you can run the full set, which is a real friction point for someone just evaluating the approach. There's no evaluation harness — you can run the notebooks but there's nothing to tell you whether retrieval actually got better after you added multi-querying. Last push was November 2025 and LangChain's API surface moves fast, so some cells are likely already broken against current package versions.