This repository curates resources for integrating Retrieval-Augmented Generation (RAG) with reasoning capabilities in Large Language Models (LLMs) and agents. It targets researchers and practitioners aiming to build more sophisticated AI systems that combine knowledge retrieval with logical processing, offering a structured overview of papers, frameworks, and benchmarks in this rapidly evolving field.

How It Works

The project categorizes advancements into "Reasoning-Enhanced RAG," "RAG-Enhanced Reasoning," and "Synergized RAG-Reasoning Systems." This taxonomy, derived from a featured survey paper, provides a framework for understanding how reasoning techniques improve RAG (e.g., retrieval optimization, generation enhancement) and how RAG enhances reasoning (e.g., external knowledge retrieval, tool use). Synergized systems are highlighted for their iterative, mutually enhancing integration of both paradigms.

Quick Start & Requirements

This repository is a curated list of research papers and implementations, not a runnable software package. It requires no installation. Users can explore linked papers and code repositories for specific implementations.

Highlighted Details

• Comprehensive taxonomy of RAG and Reasoning integration methods.
• Extensive list of recent academic papers (2024-2025) with links to code.
• Detailed table of benchmarks for evaluating RAG-Reasoning systems across various tasks (QA, summarization, fact-checking).
• Covers diverse reasoning workflows (Chain-based, Tree-based, Graph-based) and agentic orchestration.

Maintenance & Community

The repository is actively maintained by DavidZWZ and welcomes community contributions via pull requests or issues. It cites two key survey papers, one of which was featured in Hugging Face Daily Papers.

Licensing & Compatibility

The repository itself is not licensed as software. Individual linked papers and code repositories will have their own licenses, which users must consult for usage and compatibility.

Limitations & Caveats

This is a curated resource list, not an executable framework. Users must independently evaluate and integrate the linked papers and code. The rapid pace of research means the field is constantly evolving, and specific implementations may have their own limitations.