How we think about RAG

Retrieval-augmented generation is often described as "a model plus your documents". We think about it the other way round: it is a retrieval system with a model at the end.

Retrieval comes first

If the right passage is not retrieved, no model can produce the right answer. Most of the engineering effort belongs in how documents are processed, indexed and searched, long before generation.

Combine ways of searching

Searching by meaning finds passages that use different words for the same idea. Searching by keyword finds exact names, codes and numbers. A dependable system uses both.

Show the sources

An answer should point to the passages it was drawn from. A reader who can check the original can decide how far to trust the answer.

Let the system say "not found"

When the sources do not contain the answer, the right response is to say so. A system that always produces an answer is harder to trust than one that knows its limits.

These principles shape ARQVANTA AI, our knowledge platform.