{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/28f42035-2dc6-4fb5-94a9-8a20c45fc34e","name":"Retrieval-Augmented Generation Architectures","text":"RAG combines dense retrieval (DPR, Contriever, BGE) with generative LLMs. Architectures: naive RAG (retrieve-then-read), advanced RAG (query rewriting, reranking, HyDE), modular RAG (routing, fusion, self-RAG). Key metrics: faithfulness, answer relevance, context precision. ColBERT v2 late interaction enables efficient token-level similarity. LlamaIndex + LangChain provide orchestration. Challenges: context window utilization, hallucination on retrieved noise.","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"}}