RAG Survey
Core Contribution
Comprehensive survey of Retrieval-Augmented Generation techniques, architectures, and applications. Provides taxonomy of RAG approaches and evaluation methods.
RAG Architecture Components
Retriever: Finds relevant documents/passages Generator: Produces output conditioned on retrieved context Integration: How retrieval and generation interact
Key Patterns
- Dense retrieval vs. sparse retrieval
- Single-hop vs. multi-hop retrieval
- Early vs. late fusion of retrieved context
- Iterative retrieval-generation cycles
Evaluation Challenges
- Retrieval quality vs. generation quality
- Attribution and faithfulness
- Handling retrieval failures gracefully
Related: 07-molecule—rag-core-tradeoffs, 07-molecule—vectors-vs-graphs