Retrieval-Augmented Generation
RAG is an architecture where the AI retrieves relevant documents from your knowledge base before writing an answer, so it quotes your data instead of guessing from memory.
No jargon, no math
In plain words, what is it?
RAG runs in two phases. First, indexing: every document in your knowledge base is cut into chunks, each chunk is converted into a numeric fingerprint (an embedding) and stored in a vector database. Then, query time: your question is turned into a fingerprint too, the database returns the chunks with the closest fingerprints, and those chunks are pasted into the model's prompt as reference material. The model writes the answer while reading them, so it quotes your procedures, prices and logs instead of inventing them. If retrieval finds nothing relevant, a well-built RAG system says so instead of bluffing.
- 01
Indexing. Documents are split into chunks and each chunk gets a numeric fingerprint (embedding).
- 02
Storage. Fingerprints land in a vector database built for similarity search.
- 03
Query. Your question is fingerprinted the same way.
- 04
Retrieval. The database returns the chunks whose fingerprints are closest to the question.
- 05
Generation. The chunks go into the prompt; the model answers reading them, and cites the sources.
Terms that live nearby.
Vector Database
A vector database is a database built to store numeric fingerprints of content (embeddings) and find the closest ones to any query in milliseconds, powering semantic search, recommendations and AI retrieval.
ReadEmbedding
An embedding is a list of numbers that captures the meaning of a piece of text, image or audio, so that content with similar meaning gets similar numbers and the computer can measure closeness with plain arithmetic.
ReadLarge Language Model
A large language model (LLM) is an AI system trained on massive collections of text to predict the next token, which gives it a general ability to write, summarise, translate and reason without being programmed for each task.
Read§ 04
Want this running in your stack?
We design and ship AI systems with permissions, logs and human approval. It starts with a 30-minute scoping call.