Pipe(line) Dreams: Fully Automated End-to-End Analysis and Visualization

We exploit large language models (LLMs) to automate the end-toend process of descriptive analytics and visualization. A user simply declares who they are and provides their data set. Our tool LLM4Vis sets analysis goals or metrics, generates code to process and analyze the data, visualizes the results and interprets the visualization to summarize key takeaways for our user. We examine the power of LLMs in democratizing data science for the non-technical user and in handling rich, multimodal data sets. We also explore LLM4Vis's limitations, opportunities for human-in-the-loop interventions, and challenges to measuring and improving the robustness and the utility of LLM-generated end-to-end data analysis pipelines.