Do RAG and fine-tuning work together?

Often. A model can be tuned for a domain's style or task format while retrieving current, permission-filtered facts at answer time.

DECISION TOOL · FREE TO USE

RAG vs fine-tuning decision tool

Diagnose what is failing, then choose retrieval, prompting, fine-tuning or neither.

THE WORKING MODELknowledge → retrieve · behaviour → tune

Follow the decision tree

Question 1

Is the failure about knowledge: facts the model does not have, facts that change, or answers that must cite a source?

Retrieval changes what the model can consult at answer time. Fine-tuning changes how it behaves. Start by naming which one is broken.

1. Is the failure about knowledge: facts the model does not have, facts that change, or answers that must cite a source?

FROM THE RESULT TO THE DECISION

How to reason about it

Retrieval changes what a model can consult when it answers. Fine-tuning changes how it behaves. Most arguments about which to use are really arguments about what is failing, held before anyone has looked.

The tool asks the questions in the order a strong candidate would, starting with the diagnosis and trying the cheap fix before the expensive one.

Never fine-tune as a substitute for permissions or for facts that change. Use retrieval for knowledge, prompting and then tuning for behaviour, and often both.

Questions about this tool

It can make a model recall some of them, but badly: it cannot cite them, cannot enforce who may see them, and goes stale at the next edit. Retrieval is the right tool for documents.