Agentic Corrective RAG (CRAG)

Project Overview

This project showcases a small and simple implementation of Corrective Retrieval Augmented Generation (CRAG) utilizing the 🦙 llama-agents 🤖 framework. The objective of this project is to enhance the robustness and accuracy of AI-generated responses by correcting potential retrieval errors in the standard Retrieval-Augmented Generation (RAG) process.

What is CRAG?

Corrective Retrieval Augmented Generation (CRAG) is an advanced AI technique designed to address the common issue of irrelevant or incorrect document retrieval in standard RAG systems. CRAG introduces a corrective mechanism that evaluates the relevance of retrieved documents and takes appropriate actions to ensure the generation of accurate and reliable information.

The core components of CRAG include:

Retrieval Evaluator: Assesses the quality and relevance of retrieved documents.
Confidence-Based Actions: Determines the confidence level in the retrieved documents and triggers corrective actions as needed.
Knowledge Refinement: Refines and composes retrieved information to improve relevance and accuracy.
Web Search Integration: Performs web searches to gather additional information when necessary.

Getting Started

Installation

Clone the repository:

git clone https://github.com/Kirouane-Ayoub/Agentic-Corrective-RAG.git
cd Agentic-Corrective-RAG

Install the required packages:
```
pip install -r requirements.txt
```

Configuration

Embedding and llm models configurations are managed in src/settings.py. Adjust the settings according to your environment and requirements.
Create a .env file with the following content:
```
CO_API_KEY=...
```
Move your data files (pdf, csv, md, etc.) to the src/data_source folder .

Run the agent :

To run the agent locally use this command :
```
python src/local_agent.py
```

Kirouane-Ayoub / Agentic-Corrective-RAG