Context Engineering for RAG : The Four Typed Inputs Behind Every RAG Answer

Context engineering is the advanced discipline of assembling all necessary information into an LLM's context window, moving beyond simple prompt tuning to a more architectural approach. This practice involves breaking a RAG system into four "bricks": parsing, question parsing, retrieval, and generation. Each brick is responsible for emitting a specific, typed piece of information, such as structured document metadata or a parsed user query. These distinct outputs are then methodically threaded together into a single, stable LLM call, combining a fixed system prompt with the dynamically generated context.