RAG Architecture Diagram: the Boxes and What Each Arrow Costs
Short answer: A RAG architecture diagram is five boxes and the arrows that join them - an index you build, a retriever that turns a query into candidates, a reranker that orders them, a context builder that packs what the model may read, and a loop controller that decides whether any of it runs again. Every arrow carries a payload, and every payload is paid for twice: once when it is moved, once…
A RAG architecture diagram consists of five boxes and six arrows, each carrying a payload that incurs costs. These boxes include an index, retriever, reranker, context builder, and loop controller. The index dictates what the retriever can find, while the retriever transforms a query into candidate passages. The reranker orders these candidates, and the context builder determines what the model ultimately reads.
The loop controller decides whether the entire process repeats. Key takeaways include that the loop controller is where costs accumulate, and the context builder is the only box whose effect can be measured within a single request.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
