Natural Language Query Processing and on Premise Routing System for Secure Data Access

Resumen

Many times the enterprise information is shared across relational databases, internal APIs, document stores and resource portals. Without these, users must have the information related to the location and query language of each source. This paper incorporates the natural language query processing and on premise routing system which provides a single interface during directing the requests to source appropriate handlers. The methodology is designed which combines a React/Vite frontend, a FastAPI backend, PostgreSQL, pgvector based retrieval, and locally hosted large language model components via Ollama and orchestration libraries. A routing layer differentiates SQL, API, document, resource, and general questions. As security-oriented controls, authentication, request validation, controlled query execution, and logging are involved. This work documents the interface and workflow implementation having a limitation of provision of a reproducible benchmark dataset or quantitative accuracy or latency results. So this work covers functional design and testing without claiming unmeasured performance gains.