Question
Clearly sourced
Expert knowledge for digital decisions
What is RAG (Retrieval Augmented Generation)?
Short answer
RAG means that a language model looks up in your own documents before answering and only answers based on this. The model doesn't need to know anything – it just needs to be able to read. Exactly this reduces made-up answers and makes every statement traceable back to a source.
How it works
- Your documents are split into sections and made searchable.
- When asked a question, the system identifies the relevant sections.
- These sections are passed to the model together with the question.
- The model answers only based on these sections and cites the source.
Why it's better than a trained model
- Up-to-date. New documents take effect immediately; retraining would be required for a trained model.
- Traceable. Every statement can be traced back to a source.
- Controllable. Rights can be assigned per document – users who are not allowed to see a document also receive no answer from it.
- Cost-effective. No training, no specialized model required.
Limitations
The quality depends almost entirely on the search, not the model. If the wrong sections are found, even an excellent model will provide incorrect answers. This means the effort lies in preparation: meaningful section boundaries, good metadata, clean documents.
A second challenge are conflicting documents. If three versions of a guideline are stored in the archive, the system cites one of them. A document collection without a clear validity status delivers unreliable answers – regardless of the technology.
Typical Use Cases
Internal knowledge search, support based on own manuals, contract inquiries, research in project documents.
Key facts
- Core
- Search in own documents first, then answer
- Quality depends on
- the search, not the model
- Prerequisite
- Documents with identifiable validity status
Sources
All external claims are backed by traceable sources.-
01
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks Lewis et al. / arXiv