Structured AI Agents for Reliable Visualization Report Generation

Abstract:

We introduce Structured AI Agents to tackle instability and hallucination in AI agent–based data visualization. AI agents promise faster, more accessible visualization reporting, but current methods remain inconsistent and struggle to deliver reliable results. We set multiple sub-tasks and form a structured pipeline, where each sub-task unit produces standardized JSON outputs encoding executable and deterministic code. Our approach enforces correctness in quantitative results, reduces hallucinations, and ensures reliable visualizations. In our final report, we demonstrate that the modular and verifiable outputs achieve strong reliability and maintain accuracy in visualization report generation.