AI Use Cases
RAG (Retrieval-Augmented Generation)
Giving a language model relevant passages from your own documents when it answers, so the answer draws on your content and can cite the passages it used, which makes it easier to check.
Part of Cloud, Data & AI
Retrieval-augmented generation (RAG) finds passages relevant to a question in a controlled document set and passes them to a language model as context. The model is instructed to answer from those passages.
Why it is used
On its own, a language model answers from training data that does not include your internal documents and may be out of date. Grounding reduces invented answers and lets the system show which passages it used, so users can check statements against the source. It also avoids retraining: when a document changes, the new version is re-indexed (split into passages and re-embedded), and later answers draw on it.
Its limits
Grounding reduces errors without removing them. The model can still misread a passage, mix in what it learned in training, or cite a passage that does not support the claim. Answers still need human review where mistakes matter, and RAG will repeat outdated content if the source documents are not kept current.
What decides whether it works
Retrieval quality usually matters more than the choice of model. If the right passage is not retrieved, the answer rests on the wrong passages or on the model's training. Document preparation, chunking and search tuning make up most of the work. Permissions must be enforced at retrieval, so the model only receives passages the user may open.