Self-correcting Agentic RAG

This project presents a Retrieval-Augmented Generation system that intelligently combines multi-step document retrieval, dynamic query rewriting, document relevance grading, and web search augmentation. Designed as a self-correcting agent, it iteratively refines queries and supplements retrieved knowledge with real-time web data when existing documents fall short—delivering precise, contextually relevant answers to user questions.

---

Features

• Document Loading and Preprocessing: Load PDFs and split them into chunks for retrieval.
• Vector Store Initialization: Embed documents using OpenAI embeddings and index them for similarity search.
• Ensemble Retriever: Combines parent document retriever and keyword-based retriever for better results.
• Document Grading: Scores retrieved documents by relevance to the question and filters out low-quality ones.
• Query Rewriting: Improves user queries for better retrieval when documents are insufficient.
• Web Search Integration: Adds recent and relevant web search results when the retrieved documents are not relevant.
• RAG Chain for Generation: Generates answers from retrieved documents and the rewritten query.
• Flexible StateGraph Workflow: Controls the multi-step process from retrieval to final generation with conditional branching.

---

Worflow

Setup Instructions

Requirements

• Python 3.8+
• OpenAI API Key
• Required Python packages:
  • openai
  • langchain
  • some_vectorstore_library (your vector store choice, e.g., Chromadb, FAISS)
  • tavily or your web search API client

Installation:

1. Clone the Repository:

git clone https://github.com/razamehar/Self-Correcting-Agentic-RAG-Corrective-RAG.git
cd Self-Correcting-Agentic-RAG-Corrective-RAG

2. Install dependencies:

pip install -r requirements.txt

Contact

For any questions or clarifications, please contact Raza Mehar at [raza.mehar@gmail.com].