RAG (Retrieval Augemented Generation)

• This concept was termed by facebook.

• Basic LLM (Prompt/Chat interface)
  
  

• Most of the LLMs are trained on publicly available information (Website, Books, articles …..)

I want LLM to generate responses on my organization data

• Building a LLM: Costly
• RAG’s kind of solve this problem
• RAG Conceptually has two layer
  • Retriever
  • Generator: This is an existing LLM (Model)
    
• Since Retriever needs to parse the natural language
  

Technical Aspects

• Building a Vector Embedding model or Vector Database with my organization data indexed. Picking right Vector Database for the purpose

• Performing similarity searches or any other types of searches with Vector DB

• Interacting with LLM to get the response

LLM Interactions

• Direct Interactions
• InDirect Communiations using frameworks like LangChain

Integrating Existing Application with RAG

• Popular Approaches
  • REST API
  • Graph QL
• Building:
  • Fast API

Docker Desktop

• Refer Here for docker desktop