VizGenie: Toward Self-Refining, Domain-Aware Workflows for Next-Generation Scientific Visualization

Area of Science:

• Scientific Visualization
• Computational Science
• Data Analysis

Background:

• Traditional scientific visualization tools have limitations in handling complex, high-dimensional data.
• Manual script generation for advanced visualizations is time-consuming and requires specialized expertise.

Purpose of the Study:

• To introduce VizGenie, an agentic framework that leverages large language models (LLMs) to automate and enhance scientific visualization.
• To enable on-demand generation of visualization scripts and facilitate intuitive, natural language-based data exploration.

Main Methods:

• VizGenie orchestrates domain-specific modules and dynamically generates new visualization scripts using LLMs.
• It employs natural language processing and visual question answering (VQA) for high-level query interpretation.
• Retrieval-Augmented Generation (RAG) ensures reliability and provenance tracking.

Main Results:

• Automated generation and validation of visualization scripts, expanding system capabilities.
• Significant reduction in cognitive overhead for iterative visualization tasks on complex datasets.
• Successful interpretation of feature-based queries and interactive exploration via VQA.

Conclusions:

• VizGenie offers a sustainable, continuously evolving platform for scientific visualization.
• It accelerates insight generation and supports reproducible research by integrating LLM flexibility with curated tools.
• The framework enhances feature-centric exploration and adapts to user interactions.