Retrieval Augmented Generation

A method for augmenting an AI prompt with personally¹ contextualised data that is unknown to a Large Language Model. Documents are converted into vectors which are queried by a semantic search and added to the AI prompt before being sent to the LLM.

flowchart TD
	subgraph Preparation
	    Document@{ shape: docs, label: "Documents, Images, etc." } -->|chunks| model(Embedding Model)
	    model-->|vectors| vectorDB(Vector Database)
	end
    subgraph "User Prompt"
        Input@{ shape: lean-r, label: "User Prompt" }
        Input -->|semantic search| vectorDB
        vectorDB -->|extends| Context(Contextualised Prompt)
        Context --> LLM
    end

Long context prompting is an alternative to Retrieval Augmented Generation

Source: Is RAG Still Needed? Choosing the Best Approach for LLMs - YouTube

Footnotes

1. Personal as in not publicly known outside of an individual, family or business. ↩