architecture
RAG
Retrieval-Augmented Generation
Giving AI a searchable memory of your documents
Reading level
RAG systems combine a retrieval step (semantic search over a vector database) with LLM generation. Documents are split into chunks, embedded into vector space, and stored. At query time, the question is embedded, similar chunks are retrieved via cosine similarity, and injected into the LLM's context window as grounding evidence. This reduces hallucination and allows knowledge updates without retraining.
Real-world example
A company builds a chatbot that answers employee HR questions. They load the HR handbook into a RAG system. When someone asks 'How many sick days do I get?' the system finds the relevant handbook section and uses it to answer accurately — rather than guessing.