RAG

Retrieval-Augmented Generation (RAG) was introduced by Meta AI researchers in 2020 as a way to combine the generative power of language models with the precision of information retrieval. Instead of relying on knowledge baked into model weights during training, a RAG system retrieves relevant documents at inference time and gives them to the model as context.

A RAG pipeline typically works as follows: embed the user's query into a vector, search a vector database for the most similar document chunks, retrieve those chunks, include them in the prompt as context, and ask the model to answer the query based on the retrieved information. This grounds the model's response in actual sources and enables up-to-date knowledge.

RAG has become the default architecture for enterprise AI applications because it solves the training cutoff problem, enables private document Q&A, reduces hallucinations, and provides source citations. Its main weaknesses are retrieval quality (if the wrong documents are fetched, the answer suffers) and chunk fragmentation (important context may be split across chunks).