how the three RAG architectures differ in the way they retrieve information

Published 2026-08-20 in AI Agents

1. Standard RAG — "Find similar information" The left side shows traditional RAG. Flow: Query → Embedding Model → Vector DB → Context → LLM → Final Response Here is what happens: The user asks a query . Documents are converted into embeddings using an embedding model. These embeddings are stored as vectors in a Vector Database. The query is also converted into an embedding. The system performs similarity search against the vectors. The most relevant chunks are retrieved as context . The LLM uses that context to generate the final response . Example Suppose your documents contain: "AWS Glue processes customer data." The user asks: "What processes customer data?" The vector search finds the relevant chunk because the meaning of the query is very similar to the document. Best for: Simple factual questions Semantic search Document Q&A FAQ systems Single-hop retrieval Main limitation: It primarily asks: "What chunks are most similar to my query?" It doesn't naturally model explicit relationships between different pieces of information. 2. Graph RAG — "Find connected information" The middle section introduces a knowledge graph. Flow: Documents → LLM Graph Generator → Entities &…

More AI Agents articles · All collections · Practice challenges