Retrieval-Augmented Architecture
A model design that builds document lookup into its structure, so it fetches facts as part of thinking, not as an afterthought.
Think of it like
Like a lawyer whose desk is wired to the law library — pulling the right case mid-sentence rather than trusting memory.
Example
A support assistant that queries your help-center database on every question, then answers from what it found, is retrieval-augmented by design.
How it actually works
Rather than cramming all knowledge into weights, these models pair a generator with a retriever that pulls relevant text at run time, keeping knowledge fresh and updatable. It’s the architectural home of RAG. The hard parts are retrieving the right chunks and blending them cleanly into generation — bad retrieval quietly poisons good generation.
For product teams
How you keep a model current and grounded without retraining it.
For engineers
An architecture coupling a parametric generator with a non-parametric retrieval component at inference.
Related
- RAG — The technique it’s built around.
- Non-parametric memory — The knowledge store it queries.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome