RAG & Retrieval Vocabulary

Retrieval-Augmented Generation; enhancing models with external knowledge

“RAG enables the chatbot to answer questions about private company data.”

Origin: Acronym: Retrieval-Augmented Generation (Lewis et al., 2020)

a database optimized for storing and querying high-dimensional embeddings

“We use a vector database to perform semantic search on millions of documents.”

Origin: Latin vector `carrier` + database

searching by meaning rather than exact keyword matching

“Semantic search finds 'canine' when you search for 'dog'.”

Origin: Greek semantikos `significant` + search

combining keyword search and vector search for better accuracy

“Hybrid search catches both exact part numbers and conceptual queries.”

Origin: Latin hybrida `offspring of a tame sow and wild boar`

re-ordering search results using a more precise model

“A cross-encoder allows for reranking the top 50 results for better precision.”

Origin: re- `again` + rank `row, line`

splitting text into smaller segments for embedding

“Semantic chunking respects sentence boundaries better than fixed-size chunking.”

Origin: English chunk `thick piece` + -ing

inserting retrieved information dynamically into the prompt

“Context injection provides the model with the necessary facts to answer.”

Origin: Latin contextus + injectio `throwing in`