Composable Agentsfor Visual Reporting

A prototype autonomous agentic system for steerable, explainable, and collaborative analytics, exploring human–AI partnership in visualization for the generative AI era. Workshop Challenge submission for IEEE VISxGenAI 2025.

📊 Explore the Gallery

🏗️ Curious About the Design?

🧩

Composability & Modularity

We avoid brittle, monolithic models by externalizing logic to deterministic modules, leveraging the rule-based system Draco for principled, white-box autonomous visualization design.

🔍

Explainability & Trust

Every insight is accompanied by deep, executable provenance via Marimo notebooks, while all agent interactions are captured using Langfuse, making the system's reasoning transparent and its data transformations traceable.

🧭

Interactive Exploration

Powered by Mosaic, reports are not just static documents. Readers can go beyond the AI's narrative, asking their own questions via cross-filtering and direct data interaction.

🎯

Granularity & Agency

Deconstructed artifacts enable surgical modifications to any part of the report without full regenerations. Built with Observable Notebook 2.0 for source code transparency and AI-assisted editing.