Secure Natural Language Querying Across Distributed Data with RAG and Reverse Engineering

As enterprises store growing volumes of data across disparate systems and silos, enabling intuitive and secure querying becomes a major challenge—especially when metadata is sparse or user queries are context-driven. This whitepaper proposes a Retrieval-Augmented Generation (RAG)-based approach combined with reverse-engineering techniques to securely handle natural language queries. By exposing only metadata to LLMs and dynamically mapping user queries to actual file locations and data columns, the system preserves data security while delivering accurate, context-aware responses (Ni et al., 2025