Peter W. J. Staar is a Principal Research Staff Member, Master Inventor, and Manager of the AI for Knowledge group at IBM Research Europe – Zurich. He leads research and engineering at the intersection of generative AI, document intelligence, and large-scale knowledge extraction, with a focus on enabling AI systems to consume and reason over complex enterprise and scientific documents. His work spans document understanding, retrieval-augmented generation (RAG), multimodal AI, and knowledge graph construction.
Peter is the technical lead and Technical Advisory Committee (TAC) Chair of the open-source Docling project, one of the leading platforms for high-fidelity document conversion and understanding for large language models. He also chairs the TAC for Doclang, an emerging open-source initiative focused on structured document representation and agentic document workflows. Together, these projects form the foundation of a rapidly growing ecosystem for AI-native document processing adopted by researchers, enterprises, and the open-source community.
Peter joined IBM Research – Zurich in 2014 as a postdoctoral researcher after first working at IBM as a summer student in 2006. Over the past decade he has led the evolution of IBM’s document AI platform—from the original Corpus Conversion Service and Deep Search platform to today’s open-source document intelligence stack powering state-of-the-art document understanding and retrieval systems.
Before joining IBM Research, Peter was a postdoctoral researcher in Theoretical Physics and the Platform for Advanced Scientific Computing (PASC) at ETH Zurich. He received his Ph.D. (2013) and M.Sc. (2009) in Theoretical Physics from ETH Zurich and his B.Sc. in Physics (cum laude) from KU Leuven, Belgium.
His research career has evolved from high-performance scientific computing to applied machine learning, document AI, and generative AI. He has authored more than 100 scientific publications spanning high-performance computing, graph analytics, natural language processing, computer vision, multimodal document understanding, and large language models. His publications have accumulated nearly 3,000 citations, reflecting contributions across both scientific computing and AI.
Peter has received numerous distinctions throughout his career. He was twice a finalist for the ACM Gordon Bell Prize, winning the award in 2015 for pioneering work on extreme-scale simulations of Earth’s mantle after an earlier finalist nomination in 2013 for large-scale quantum many-body simulations. Additional recognitions include the IPDPS 2016 Best Paper Award for scalable graph analytics and the IAAI 2021 Innovative Application Award for pioneering machine learning techniques for robust PDF document conversion. More recently, his research has contributed to influential open datasets, document understanding benchmarks, and state-of-the-art models that underpin the Docling ecosystem and modern enterprise document AI.
