<![CDATA[The deployment of large language models (LLMs) in enterprise environments faces two fundamental challenges: single-search strategies are unable to effectively handle the structural heterogeneity of operational queries, and LLM generation without external grounding is prone to producing hallucinations on structured domain data. This research aims to design and evaluate a Hybrid Semantic Retrieval-Augmented Generation (Hybrid RAG) system that integrates a dual-LLM routing mechanism to address both issues simultaneously. The proposed system combines FAISS-based dense vector retrieval and sparse BM25 retrieval fused via Reciprocal Rank Fusion (RRF), accompanied by a routing module that directs domain-specific queries to a context-aware LLM (Grok) and general queries to a broadly knowledgeable parametric LLM (Gemini). Evaluation was conducted on 200 labeled queries using the RAGAS framework with an ablation study of six system configurations. The proposed system achieves a Faithfulness of 0.91, Answer Relevance of 0.87, Context Precision of 0.81, Context Recall of 0.86, Mean Reciprocal Rank of 0.84, and a routing accuracy of 93.5% meeting all seven set target thresholds. Ablation studies demonstrate that RRF fusion improves Context Recall by up to 30.3% compared to the dense-only baseline, while dual-LLM routing yields a 68.5% increase in Faithfulness relative to the baseline without RAG. These findings indicate that the combination of retrieval and dual-LLM routing produces a synergistic effect that cannot be achieved by either component alone, thereby providing a practical and scalable solution for structured operational data-based question-answering systems in enterprise environments.]]>
