Mind2Report: A Cognitive Deep Research Agent for Expert-Level Commercial Report Synthesis
This work addresses the challenge of generating high-quality, comprehensive, and reliable business reports from vast amounts of noisy web data to support high-stakes decision-making. The authors propose a training-free agentic workflow that emulates the cognitive process of professional business analysts. Their approach integrates fine-grained intent parsing, dynamic web retrieval, real-time information distillation, and iterative report generation, augmented by a dynamic memory mechanism to enhance the long-horizon reasoning capabilities of large language models. Evaluated on QRC-Eval—a newly curated benchmark comprising 200 real-world business tasks—the proposed method significantly outperforms state-of-the-art deep research agents from OpenAI and Gemini, achieving expert-level performance in automated business report synthesis.