Higress-RAG: A Holistic Optimization Framework for Enterprise Retrieval-Augmented Generation via Dual Hybrid Retrieval, Adaptive Routing, and CRAG

Researchers have developed Higress-RAG, a new enterprise-grade framework that addresses key challenges in Retrieval-Augmented Generation systems including low retrieval precision, hallucination, and high latency. The system introduces innovations like 50ms semantic caching, hybrid retrieval methods, and corrective evaluation to optimize the entire RAG pipeline for production use.

• →Higress-RAG addresses three major RAG challenges: low retrieval precision, high hallucination rates, and unacceptable latency for real-time applications.
• →The system implements a 50ms-latency semantic caching mechanism with dynamic thresholding for improved performance.
• →The framework uses Reciprocal Rank Fusion (RRF) to merge dense and sparse retrieval signals for better accuracy.
• →Built on Model Context Protocol (MCP), it offers a layered architecture with adaptive routing and corrective evaluation.
• →Experimental results show the system provides a scalable, hallucination-resistant solution for enterprise AI deployment.