glossary
RAG Architecture
Context infrastructure: sources, event-driven indexing, retrieval with re-ranking, context supply and quality evaluation.
RAG architecture (retrieval-augmented generation) is context infrastructure: normalizing and versioning sources, indexing on change events, hybrid retrieval with re-ranking, supplying the model the minimum sufficient context, and continuously evaluating answer relevance and groundedness.
How it differs from the naive version: “a vector database with a prompt” works in a demo because the data is small, fresh and selected to match the questions. In production every one of those assumptions breaks: the index goes stale, retrieval gets noisy, context is unbounded — and the system starts answering confidently wrong.
Why it matters: answer quality is determined by which context, and how much of it, reached the model — not by the size of the model. Without quality evaluation and index-freshness control, RAG degrades invisibly, and it stops being trustworthy exactly when it matters most.