What does Retrieval Augmented Generation mean?
Retrieval Augmented Generation (RAG) combines information retrieval with a language model’s response. First, the application searches for content relevant to the question. The model then receives that content as context for its answer. This makes it possible, for example, to include approved company knowledge.
AN EXAMPLE IN PRACTICE
Where you might encounter it
Your team asks how to set up a product. The application searches for relevant passages in the current manual and formulates an answer with references. Someone can look up the passages used directly.
Why does this matter for your business?
Current information can be used without retraining a model every time something changes. Good results depend on suitable sources and an appropriate search process.
Does RAG prevent every incorrect answer?
No. The search may return unsuitable passages, and the language model may combine information incorrectly. So check the search and the answer separately. A knowledge assistant should clearly identify missing sources.
Frequently asked questions
Do my data have to be used to train the model?
RAG passes relevant content to the model for each request. This is different from training. You also need to clarify how the service you use processes and stores inputs.
Which documents are suitable?
Examples include reviewed manuals, product information, and internal instructions. Outdated versions, conflicting content, and poor-quality scans should be cleaned up beforehand.
What to keep in mind
- Keep documents up to date and traceable.
- Take permissions into account during the search itself.
- Show sources and test answers with sample questions.
Further sources
From NetzTurm Decode. Clear explanations for your business.