RAG from zero to Graph RAG, one chapter at a time

What it does This repository is essentially an open-source textbook for retrieval-augmented generation. It walks developers through the full stack—data chunking, vector and multimodal embeddings, Milvus and Neo4j indexing, hybrid retrieval, query rewriting, Text2SQL, generation formatting, and system evaluation—pairing each concept with code examples. The syllabus culminates in two project tracks: a standard RAG pipeline and a Graph RAG optimization variant.

The interesting bit The maintainers treat RAG as an engineering discipline rather than a quick LangChain demo. They dedicate entire chapters to evaluation methodology and index optimization—topics usually glossed over in quickstarts—and use an “Extra Chapter” section to let the community submit deep dives, which keeps the curriculum from fossilizing.

Key highlights

• Ten-chapter progression from RAG basics to hybrid search, Text2SQL, and Graph RAG architecture
• Multimodal coverage including vector embeddings, Milvus, and a Jina v5-omni extra chapter
• A full evaluation chapter with tools and metrics instead of guesswork
• Community-contributed extra chapters (Neo4j, omni-embedding) vetted for depth
• Bilingual support: Chinese primary content with an English README available

Caveats

• Chapter 10 (the second end-to-end project) is marked “in planning,” so the course is not yet complete
• The Jina v5-omni multimodal embedding extra chapter is labeled “optimizing,” indicating it is still rough
• Code and explanations are Chinese-first; English speakers will need to rely on the separate English README

Verdict A strong bookmark for Chinese-speaking Python developers who want a structured, syllabus-style path from RAG basics to production concerns. Look elsewhere if you need a drop-in framework or a fully finished, packaged library.