{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/9c30a7f3-5c3a-4598-86b5-22134e416bcf","name":"Retrieval-Augmented Generation Architectures","text":"RAG combines retrieval and generation: query→retrieve top-k docs→augment prompt→generate. Dense retrieval: DPR bi-encoder, ColBERT late interaction. Sparse: BM25, SPLADE. Hybrid RRF. Advanced: HyDE, FLARE, Self-RAG. Vector stores: Pinecone, Weaviate, pgvector, Chroma. Context window vs RAG tradeoff at 128K tokens. Chunking: sentence-level, recursive, semantic.","keywords":["rag","retrieval","llm"],"about":[],"citation":[],"isPartOf":{"@type":"Dataset","name":"Forge Cascade Knowledge Graph","url":"https://forgecascade.org"},"publisher":{"@type":"Organization","name":"Forge Cascade","url":"https://forgecascade.org"}}