Skip to content

peter-gy/VISxGenAI-2025

Repository files navigation

Agentic Visual Reporting IEEE VIS 25 Logo

A composable agentic system for automated visual data reporting, implementing a Human-AI Partnership model for transparent and steerable data analysis.

Overview

This prototype system demonstrates a multi-agent architecture that strategically externalizes logic from LLMs to deterministic modules, leveraging Draco for principled visualization design. The system delivers dual outputs:

  • Interactive Observable reports with Mosaic for reader exploration
  • Executable Marimo notebooks for deep, analyst-facing traceability

For complete details, see our whitepaper.

Architecture

System Architecture

The system consists of 11 specialized agents orchestrated in a multi-stage pipeline:

  1. Data Understanding: Field Refiner → Dataset Describer → Field Expander → Dataset Profiler
  2. Analysis: Insight Planner → Dataset Deriver → Dataset Publisher
  3. Visualization: Dataset Visualizer (powered by Draco)
  4. Reporting: Report Narrator → Dataset Reporter

Read more in our whitepaper.

Gallery

Explore example reports generated by the system in the gallery.

Getting Started

Ready to go deeper? Check out our Development Guide for complete setup instructions, API configuration, and usage examples.

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •