Dongsheng Wang, PhD

Incoming Lead Applied Scientist @ Thomson Reuters Lab AI Quant @ Verition • ex-J.P. Morgan AI Research Lead

Biography

I build high-impact AI platforms and foundation models at the intersection of deep learning, LLM, Agent and quantitative systems. At Verition Fund Management's AI Lab, I built the firm's core AI platform from 0 to 1, and in October 2026, I will be joining Thomson Reuters Lab as Lead Applied Scientist. Previously, I was AI Research Lead at J.P. Morgan AI Research in London, where I initiated and led the flagship DocLLM project.

I received my Ph.D. in Computer Science (research on Deep Learning, NLP & IR) from the University of Copenhagen (supervised by Prof. Christina Lioma and Prof. Jakob Grue Simonsen, as part of the EU Horizon 2020 Marie Skłodowska-Curie program). Earlier, I received my M.S. in Computer Science from Korea University, and gained foundational AI research and engineering experience at Tencent AI and the Chinese Academy of Sciences (CAS).

Core Expertise & Engineering Impact

Document AI & Foundation Models — Pioneering architect of layout-aware generative models (DocLLM, ACL 2024) and graph-augmented models (DocGraphLM, SIGIR 2023) for complex enterprise documents.
Generative AI Platforms & Agents — Architect of production-grade AI platforms, Hierarchical Deep-RAG agent orchestration, Model Context Protocol (MCP) integrations, and semantic feature pipelines in quantitative finance.
Knowledge Representation & Tech Vision — Built large-scale Knowledge Graphs at CAS (CASIA-KB and Linked-Brain-Data in collaboration with EPFL's Human Brain Project), before pivoting to Neural NLP and Agentic LLM architectures. Author of 5 Reasons Knowledge Graph Will Never Bloom (2021).

Featured Publications & Patents

[ACL 2024] DocLLM: A layout-aware generative language model for multimodal document understanding. (Main Conference)
[SIGIR 2023] DocGraphLM: Documental Graph Language Model for Information Extraction.
[ACL 2025] CoCoLex: Confidence-guided Copy-based Decoding for Grounded Legal Text Generation.
[US Patent 2023] System and method for layout-aware generative pretraining for visually rich document understanding. (J.P. Morgan)
[PhD Thesis] Semantic Representation and Inference for Natural Language Processing.

Note: For a full publication list (1,000+ citations), please visit my Google Scholar.

Selected Keynotes & Talks

Invited Speaker (2024): UCL TechSocial Series — Document AI and Large Language Models.
Invited Keynote (2023): Shanghai Library Lecture — Large Language Models and Knowledge Graphs.

Honors & Awards

1st Place Winner — CLEF 2018 Lab Task on Automatic Fact-Checking.
2nd Place Winner — NLP&CC 2013 Shared Task on Chinese Entity Linking.