What is Retrieval-Augment Generation (RAG) and Why is it Needed?

Retriever-Augmented Generation (RAG) architecture was conceived to overcome fundamental limitations LLM-only generation approach. Retraining of LLMs for domain and business specific cases is still extremely expensive undertaking. Also there is fundamental competition between the need for LLM to be generally usable (have common knowledge) and specifically usable (have domain or business case knowledge). So RAG is here to help overcome LLM limitations and enable more engineering based solutions.

Retriever-Augmented Generation core components are:

1. Indexing pipeline - system that processes business or domain specific documents and stores them into knowledge base;
2. Knowledge base - database that stores domain or business specific information to be used as context for user prompts
3. Retriever - a system that based on user query tries to fetch most relevant domain or business specific information
4. LLM - large language model that receives as prompt user query with additional relevant information collected by retriever

​

​