Towards Verifiable Multimodal Deep Research: A Multi-Agent Harness for Interleaved Report Generation

Chenghao Zhang, Guanting Dong, Yufan Liu, Tong Zhao, Zhicheng Dou

arXiv:2605.29861 · 2026-05-30 공개 · arXiv · PDF

deep-research interleaved-generation cross-modal-consistency verifiable-research multimodal-reports visual-working-memory ptah ptah-eval

Abstract

Large Language Models (LLMs) have advanced autonomous agents from deep search, which retrieves concise factual answers, to deep research, which synthesizes scattered evidence into long-form reports. However, verifiable multimodal deep research remains challenging due to open-ended synthesis without deterministic ground truth and the need to interleave textual arguments with visual evidence. We propose Ptah, a multi-agent harness for interleaved report generation. Ptah orchestrates the lifecycle from user query to rendered web report through planning, research, and writing stages, where specialized agents construct visual-aware plans, collect claim-grounded evidence, maintain source-aligned images in a Visual Working Memory, and compose reports through declarative multimodal tool use. A verifier agent serves as the harness's acceptance function, enforcing factual grounding, citation fidelity, and cross-modal consistency throughout the workflow. We further introduce PtahEval, an evaluation protocol that augments existing benchmarks with image-level and presentation-level assessments. Experiments on deep research benchmarks show that Ptah produces more reliable, visually informative, and usable human-facing multimodal reports than strong baselines.

한국어 요약

📋 한 줄 요약

**[Deep Research / Multimodal Multi-Agent]** Ptah가 visual-aware 계획·claim-grounded 수집·source-aligned 이미지 working memory·verifier agent로 텍스트·시각 interleaved 보고서 생성, PtahEval로 image·presentation 수준 평가 추가.

🎯 핵심 기여도

💡 핵심 아이디어

Verifiable multimodal deep research는 텍스트·시각 evidence가 cross-modal로 일관 grounded돼야 하며, planning·research·writing 단계마다 verifier가 acceptance function으로 작동하는 multi-agent harness와 Visual Working Memory의 source-alignment가 핵심이다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Open-ended multimodal synthesis의 verifiability 문제 해결, Visual Working Memory·verifier acceptance function의 일반화 가능한 패턴, PtahEval로 평가 표준 확장. **한계**: 다단계 multi-agent의 latency·cost, verifier acceptance function의 false negative 가능성, 평가 augment의 인간 annotator 의존 가능성.

🚀 실용적 활용