A Closed-Domain Natural Language Database Query System by LLMs

In this study, we have proposed an innovative database query system integrated with natural language processing capabilities of large language models (LLMs); in which, the privacy and public access problems are addressed by a local deployment. To confront the challenges of providing privacy protection and access controls commonly available in traditional database system, our system operates in a closed network environment. By combining prompt engineering with structured databases, it facilitates non-technical users with efficient complex data retrieval using natural language. The front-end interface, built with Streamlit, enhances usability, while Docker-based containerization ensures consistent deployment and maintenance. Performance evaluations using multiple benchmark datasets demonstrate that the proposed prompt strategies significantly improve SQL generation accuracy, confirming the system's practicality and flexibility across diverse application scenarios.