What it does

All-in-RAG is a comprehensive Chinese-language tutorial for building retrieval-augmented generation systems. It walks through the entire pipeline: data loading and text chunking, vector and multimodal embeddings, vector database construction with Milvus, hybrid retrieval, query rewriting, Text2SQL, formatted generation, and system evaluation. Two end-to-end projects are included, with an optional Graph RAG optimization track using Neo4j.

The interesting bit

Most RAG content is scattered blog posts; this attempts to be a single curriculum. The project also accepts community contributions as “Extra Chapters” — one contributor already added a Neo4j simple application, another is optimizing a multimodal omni-embedding practice with Jina v5-omni.

Key highlights

• Ten chapters progressing from “four steps to RAG” to advanced architecture
• Covers multimodal embeddings (text + image) and hybrid search (dense + sparse)
• Includes practical Milvus deployment and index optimization
• Graph RAG chapter with knowledge graph construction and intelligent query routing
• Systematic evaluation chapter with tools and metrics, often skipped in tutorials
• Extra Chapter section open to community submissions with review process

Caveats

• Primary content is in Chinese; English README exists but scope is unclear
• Chapter 10 (second project) is marked “规划中” — in planning, not yet available
• Some Extra Chapter content is noted as “优化中” (being optimized)
• Docker and basic Linux command line skills are expected prerequisites

Verdict

Worth bookmarking if you’re a Mandarin-speaking developer who needs structured RAG training rather than piecemeal Medium articles. Skip if you need mature, production-hardened frameworks or prefer English-first documentation.