Conversational A.I for Smart Exploration (C.A.S.E)

In today's data-driven world, the ability to extract meaningful insights remains a significant bottleneck, often confined to technical specialists. This paper introduces Conversational A.I. for Smart Exploration (C.A.S.E.) a revolutionary platform designed to dismantle this barrier. C.A.S.E. transforms the complex, code-heavy process of data analysis into a simple, interactive dialogue, empowering any user—from business executives to domain experts—to converse directly with their datasets. By simply asking questions in natural language, users can automatically uncover insights, generate dynamic visualizations, and even build predictive models, turning raw data into a decisive, strategic advantage without writing a single line of code.At its core, C.A.S.E. operates on a sophisticated multi-agent framework where specialized agents collaborate to orchestrate the entire workflow. The Insight Generation Module, inspired by the QUIS framework, automates the discovery of statistically significant patterns through an iterative question-generation and subspace search pipeline. For complex tasks like preprocessing and visualization, C.A.S.E. employs a hybrid "caller-or-coder" design, balancing the reliability of predefined tools with the flexibility of Large Language Model (LLM)-generated code. This architectural prowess culminates in the AutoML module, which introduces a novel supervisor-agent architecture, moving beyond rigid, static pipelines to enable adaptive, iterative optimization of model development. Our results demonstrate consistent performance gains, with C.A.S.E. achieving the highest or matching F1-score on all classification benchmarks and the lowest or matching RMSE on every regression dataset tested, including a 0.6 F1-score improvement on the Banana Quality dataset and a 5 RMSE reduction on the NYC Airbnb dataset. By integrating these components into a seamless workflow, C.A.S.E. delivers a holistic solution that truly democratizes data science.