Understanding the RAG Process

RAG Process Overview

• The RAG process consists of four steps: Retrieval, Augmentation, Generation, and Response.
• The process is designed to provide informed responses to user queries by leveraging external contextual documents.

Step 1: Retrieval

• Contextual documents are retrieved from an external dataset.
• The retrieval process is based on the similarity between the user query and the documents in the dataset.

User Query and Conversion

• A user poses a natural language query to the LLM (e.g., "Tell me about the Renaissance period").
• The query is converted into a numeric format using an embedding model, creating a vector representation.
• The embedding model used for query conversion is the same as the one used for article embedding in the ingestion phase.

Vector Comparison and Retrieval

• The query vector is compared to vectors in the knowledge base index using similarity or distance metrics (e.g., cosine similarity).
• The system retrieves the top-K documents or passages with the highest similarity to the query vector.

Remaining Steps

• Augmentation: The retrieved documents are integrated with the original input to enrich the context.
• Generation: The model generates a response based on the augmented input.
• Response: The informed response, influenced by the retrieved contextual documents, is delivered to the user.