RAG Architecture

1. What the Pipeline Actually Is

Retrieval-augmented generation is usually drawn as two boxes: a vector search and a model. The drawing is why most of these systems disappoint.

The real pipeline has more stages, and the quality of the answer is decided almost entirely by the stages before the model.

ingest  ->  chunk  ->  embed  ->  index
query   ->  rewrite  ->  retrieve  ->  rerank  ->  assemble  ->  generate

The useful mental model is that the model is not the part you are designing. It is a fixed component that turns a context window into prose