Bridging Natural Language and Databases: Intelligent Conversational System for Automated SQL Query Generation Using Natural Language Processing

Abstract

Relational databases are the foundation of modern data management systems, widely utilised across industries for storing, organising, and retrieving information. However, interacting with these databases typically requires proficiency in Structured Query Language (SQL), which limits accessibility for non-technical users. Addressing this challenge, this paper presents a multimodal AI-powered SQL chatbot that allows users to query relational databases using natural language, voice commands, and visual schema inputs. The proposed system employs a fine-tuned T5 transformer model to convert natural language statements into executable SQL queries. It further integrates Optical Character Recognition (OCR) for schema extraction and multilingual translation modules to support diverse user inputs. Unlike conventional Text-to-SQL systems, this approach emphasises user-centric design, providing both textual and graphical result visualisation for improved interpretability. Experimental evaluation was conducted using benchmark datasets such as Spider and WikiSQL, alongside custom domain-specific datasets covering HR, Sales, and Academic data. The model achieved competitive accuracy and low latency, averaging 1.5–2 seconds per query, while ablation studies confirmed the impact of OCR and multilingual modules on performance. Furthermore, a user study involving 20 participants demonstrated high usability and satisfaction ratings, validating the chatbot’s potential to make database querying more natural and inclusive. This research contributes toward the democratisation of data access by merging transformer-based NLP, multimodal interaction, and user-focused design into an integrated, practical solution.