OPTIMIZED DOMAIN-SPECIFIC KNOWLEDGE RETRIEVAL USING HYBRID RAG ARCHITECTURE

Abstract

Efficient retrieval of domain-specific knowledge is essential for improving the accuracy and reliability of intelligent systems in specialized fields such as healthcare, finance, and legal analytics. Traditional information retrieval methods often fail to capture semantic context, while large language models (LLMs) may produce responses lacking factual grounding. To address these limitations, this paper proposes an optimized domain-specific knowledge retrieval framework using a Hybrid Retrieval-Augmented Generation (RAG) architecture. The proposed system integrates dense vector retrieval with sparse keyword-based retrieval to combine semantic understanding with exact term matching. Domain-specific embedding models are used to generate meaningful vector representations, enabling efficient similarity search in a vector database. A lexical retrieval component complements this process by improving precision for specialized terminology. The retrieved results are further refined using a re-ranking mechanism to ensure contextual relevance before being passed to a generative model. The generation module produces accurate, context-aware responses grounded in retrieved knowledge, reducing hallucination and enhancing reliability. Experimental results show improved retrieval accuracy and response quality compared to traditional methods. The proposed hybrid RAG framework is scalable, efficient, and well-suited for real-world applications requiring precise domain-specific knowledge access.