RAG

Grounded generation is when an AI model answers using supplied source material rather than its own memory, ideally citing exactly which passages support each claim.

A retriever is the component of a RAG system that takes a query and returns the passages most likely to contain the answer. Its recall sets the ceiling on how good the final answer can be.

A reranker is a model that reorders a set of retrieved candidates by scoring each one directly against the query, producing much more accurate ordering than the initial vector search.

Document chunking is the process of splitting documents into smaller passages so they can be embedded, retrieved and passed to a model. Chunking decisions have more effect on RAG quality than almost any other choice.

Hybrid search combines keyword search, which excels at exact terms, with vector search, which captures meaning, then merges the two result lists into a single ranking.

Semantic search finds results based on meaning rather than exact word matching, so a search for 'can't log in' returns a password reset guide even though it shares no keywords with the query.

Vector search is the operation of finding the embeddings closest to a query vector. At scale it uses approximate nearest neighbour indexes, which trade a small amount of accuracy for very large speed gains.

A vector database is a database built to store vector embeddings and find the most similar vectors to a query quickly, usually combining approximate nearest neighbour indexes with metadata filtering.

A RAG pipeline is the end-to-end system that turns raw documents into grounded answers, covering ingestion, chunking, embedding, storage, retrieval, reranking, prompting, generation and evaluation.

Retrieval-augmented generation (RAG) is a technique where relevant documents are retrieved from a knowledge source and added to an LLM's prompt, so its answers are grounded in specific, current information rather than only its training data.