RAG

RAG, or retrieval-augmented generation, hands a model your own documents before it answers. The model then quotes your policies instead of inventing them. This tag covers what a useful knowledge base needs: clean sources, sensible chunks and real search. It also covers where the approach beats fine-tuning, and where it quietly fails. Bad documents produce confident nonsense.