This chapter covers
- Introducing RAG variants
- Knowledge graph RAG
- Multimodal RAG
- Agentic RAG
- Other RAG variants
The first part of the book introduced retrieval-augmented generation (RAG) and the core idea behind it. The second part dealt with building and evaluating basic RAG systems. Part 3 took RAG beyond the naïve approach and discussed advanced techniques and the technology stack that supports a RAG system. The last part of the book looks at more RAG patterns, and we conclude our discussion with a few best practices and some areas for further exploration.
Chapter 8 looks at some popular RAG variants. These variants adapt different stages of RAG (i.e., indexing, retrieval, augmentation, and generation) to specific use case requirements. The chapter begins by discussing the emergence of these variants and the purpose they serve. We then continue talking about three important variants that have gained prominence in applied RAG. These are knowledge-graph-enhanced, multimodal, and agentic RAG. We also briefly examine other RAG variants that significantly contribute to the evolution of RAG in practical applications. We discuss the purpose and motivation behind each variant. This chapter also breaks down the workflow, features, and technical details of the variants along with their strengths and weaknesses. For simplicity, the code for these variants is not included in this chapter but can be found in the book’s code repository.