RAG

Retrieval-Augmented Generation

Pattern where an LLM is grounded by retrieving relevant docs at inference time.

When you'd see it: Building LLM applications that need current or private knowledge, like a chatbot over your company's documents.

Why it matters: RAG grounds a language model by retrieving relevant documents at query time and feeding them in, so answers reflect your data instead of only the model's training. It is the dominant pattern for making LLMs useful on private knowledge without retraining them.

Common mistakes: Expecting RAG to fix a weak model. If retrieval surfaces the wrong documents, the model confidently answers from the wrong context. Retrieval quality is usually the bottleneck.