This paper presents a constraint-based generative user interface system that dynamically constructs analytical dashboards from raw CSV datasets using large language models (LLMs) and automated statistical profiling. The system introduces a self-healing data pipeline that detects and corrects type mismatches, temporal inconsistencies, and null values before analysis. It integrates statistical analysis, natural language insights using LLaMA 3, rule-based chatbot querying, and linear regression prediction into a unified framework. The system reduces manual effort and enables non-technical users to generate insights, visualizations, and forecasts automatically from raw data.
Srivarshini et al. (Mon,) studied this question.