Wyvern: An Agentic Framework for Generating Grounded Multimodal Reports

2026-08-14

AI Summary

This study addresses the lack of grounding and difficulty in tracking knowledge updates in generative model outputs by proposing Wyvern, a multi-agent framework. Wyvern integrates multimodal data with citation support to automatically generate grounded technical reports comprising text, figures, and tables, while incorporating an automatic claim correction mechanism to enhance reliability. Experimental results demonstrate that Wyvern outperforms baselines in figure informativeness in 87% of cases and yields more practical reports in 63%–100% of instances. Furthermore, it achieves a 2.3-fold increase in citation recall and a 1.6-fold improvement in precision. These findings indicate that Wyvern significantly enhances both the accuracy and trustworthiness of multimodal technical report generation, effectively mitigating hallucination issues through rigorous evidence integration and automated verification.

Abstract

In the current artificial intelligence-driven innovation era, the pace of knowledge growth is accelerating, and is hard to keep up with. While generative models are increasingly used to synthesize content, they often lack in information grounding. To address these peculiarities of our time, we propose Wyvern, a multi-agent framework for the automated generation of grounded, multimodal technical reports. Wyvern allows for the generation of multimodal outputs, integrating images, tables, and text with supporting references in a unified report. Additionally, a particular focus is placed on the grounding of the content, with the implementation of a claims auto-revision stage. We conduct a human evaluation study to assess the quality of our proposed framework. The results show that the figures' informativeness is perceived as superior to that of a recent baseline in 87% of cases. Furthermore, Wyvern's reports are rated as more useful than those produced by three alternative methods in 63% to 100% of instances. We also carry out automatic evaluations showing that Wyvern gains up to 2.3$\times$ in citation recall and 1.6$\times$ in citation precision with respect to the baselines.

Problem

• Multimodal Report Generation
• Information Grounding
• Generative Models
• Citation Accuracy

Innovation

• Multi-agent Framework
• Grounded Generation
• Multimodal Reports
• Claims Auto-revision
• Citation Recall