Reproduce “Learning to Route: A Rule-Driven Agent Framework for Hybrid-Source Retrieval-Augmented Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Haoyue Bai, Haoyu Wang, Shengyu Chen, Zhengzhang Chen, Lu-An Tang, Wei Cheng, Haifeng Chen, Yanjie Fu

Researchers from Arizona State University and NEC Laboratories America developed a rule-driven agent framework for Retrieval-Augmented Generation (RAG) that dynamically routes queries to relational databases, unstructured documents, or both. This framework enhances answer accuracy, improves token efficiency, and reduces routing latency by leveraging a rule-making expert agent and a path-level meta-cache.

Key insights

  • Relational databases and unstructured documents offer complementary strengths, with databases excelling in fact-centric queries and documents in open-ended or descriptive queries.
  • Intelligent routing to the most suitable knowledge source is essential to improve RAG accuracy and efficiency, as naive hybridization is often counterproductive.
  • A transparent, rule-driven routing approach offers interpretability and adaptability, outperforming black-box learned routers and providing efficient decision-making akin to human heuristics.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.