RAG : Retrieval Augmented Generation
RAG : Retrieval Augmented Generation
Now that you understand the principles of Context Engineering, let's look at one of the most widely used engineering patterns in modern AI. It is called RAG, which stands for Retrieval-Augmented Generation. Despite the technical name, the idea is surprisingly simple. Instead of expecting the LLM to know everything, we allow it to read the relevant information first, and only then generate an answer. Let's break the name into three parts. Retrieval means finding the most relevant information from an external source such as PDFs, company documents, databases or knowledge bases. Augmentation means placing only that relevant information inside the model's context window. Finally, Generation means allowing the LLM to generate its answer using both its own knowledge and the newly retrieved information. That complete sequence is called a RAG pipeline. Consider a practical example. An employee asks, "How many casual leaves do I get every year? " The answer exists inside a 200-page HR manual. Instead of forcing the LLM to memorize the entire handbook, a retrieval engine searches the document, identifies the leave policy, extracts only the relevant paragraphs and injects them into the context window. The LLM then answers based on those paragraphs. The model never searched the document. It never memorized the handbook. It simply reasoned over the information that was supplied to it. This is why RAG has become the standard approach for enterprise AI. Company policies change. Product catalogs change. Legal documents change. Medical guidelines change. Rather than retraining the LLM every time your data changes, you simply update your knowledge source. The next time someone asks a question, the retrieval system fetches the latest information before the model responds. If you build systems using LangChain, LangGraph, LlamaIndex, Haystack or n8n, you will frequently encounter terms like Retriever, Knowledge Base, Document Loader, Vector Store and RAG Pipeline. Different frameworks use different names, but they all implement the same idea. Find the right information, place it into the context window and let the LLM reason over it. One final insight is worth remembering. RAG does not eliminate hallucinations. It reduces them by grounding the model in relevant information. If the wrong document is retrieved, the answer can still be wrong. If the knowledge base is outdated, the answer will also be outdated. This is why building a good RAG system is not just an AI problem—it is also a knowledge management problem. And this brings us back to a principle you have already learned. Precision and relevance do not come from the LLM. They come from the engineer and the domain expert who decide what knowledge should be available, how it should be organized and what information should be retrieved. RAG is simply the engineering pattern that connects that knowledge to the reasoning power of an LLM.