In this paper, we present a novel hybrid framework that integrates Retrieval-Augmented Generation (RAG) and intent-based structured answers to address the challenges of enterprise-scale deployment of large-scale language model (LLM)-based chatbots, such as diverse user questions, high latency, hallucinations, and difficulties in integrating frequently updated domain-specific knowledge. The framework dynamically routes complex or ambiguous questions to the RAG pipeline while leveraging predefined high-confidence answers for efficiency. It uses a conversational context manager to maintain consistency across turn-heavy interactions, integrates feedback loops to improve intent, dynamically adjusts confidence thresholds, and expands response ranges over time. Experimental results show that the proposed framework achieves high accuracy (95%) and low latency (180ms), outperforming RAG and intent-based systems across a wide range of question types, and is positioned as a scalable and adaptable solution for enterprise-scale conversational AI applications.
