Retrieval-Augmented Generation (RAG) Implementations

This repository contains various implementations of Retrieval-Augmented Generation (RAG) techniques using LlamaIndex and OpenAI models. Each implementation explores a different approach to enhance information retrieval and generation capabilities. Different rag implementations from different frameworks, VectorDBs and models will be added regularly.

📂 RAG Types

The following types of rags are currently available in the repository:

1. vanilla-rag

   • Description: A basic implementation of RAG using vector-based retrieval.
   • Key Features:
     • Simple document retrieval using embeddings.
     • Integration with OpenAI's language models for query answering.

2. rerank-rag

   • Description: RAG implementation with a reranking mechanism to improve retrieval quality.
   • Key Features:
     • Rewriting the query for better search
     • Uses a reranker (bi-encoder --> sentence-transformers/all-MiniLM-L6-v2) to prioritize retrieved documents.
     • Enhanced accuracy for complex queries.

3. graph-rag

   • Description: Graph-based RAG implementation that leverages knowledge graphs for retrieval.
   • Key Features:
     • Builds a knowledge graph from documents.
     • Expands retrieval using graph relationships.

4. agentic-rag

   • Description: Agent-based RAG implementation with tools for advanced reasoning.
   • Key Features:
     • Uses a ReAct agent to process queries.
     • Integrates tools for document search and analysis.