Retrieval Augmented Generation

RAG connects a Large Language Model (LLM) to an external, authoritative knowledge base, retrieving specific data to ground its response and eliminate hallucinations.

Retrieval-Augmented Generation (RAG) is a critical framework: it injects external, non-parametric knowledge into the LLM generation process. The system first uses a user query to retrieve relevant document chunks from a vector database (e.g., company internal data). This retrieved context is then prepended to the original prompt, effectively grounding the LLM's output. This process significantly boosts factual accuracy, minimizes the risk of AI hallucination, and allows for continuous knowledge updates without the massive computational expense of fine-tuning the base model.

https://arxiv.org/abs/2005.11401