Empowering intramural researchers with cutting-edge AI analytical tools and data
OIR Welcomes Dr. Nic Dobbins

The Office of Intramural Research (OIR) welcomes Nic Dobbins to the Clinical Research Informatics Supporting Principal Investigators (CRISPI) team. Dobbins is a scientific and engineering expert in the intersection of large language models (LLMs) & natural language processing (NLP), cohort discovery, and biomedical databases. Prior to joining NIH, Dobbins was an Assistant Professor of Biomedical Informatics & Data Science at Johns Hopkins University, and later Head of Research Engineering at Medeloop, a San Francisco-based AI startup focused on self-service LLM-driven medical analytics.
He held additional leadership positions as Assistant Director of Education in Biomedical Informatics at Johns Hopkins University and Adjunct Professor of Biomedical Informatics at the University of Washington (UW) in Seattle.
Making accessing, exploring and analyzing clinical data faster and easier
Joining NIH, OIR and the CRISPI team, Dobbins is excited to build upon projects and the CRISPI vision for empowering intramural investigators and accelerating research efforts across NIH: “I’m thrilled and humbled to be here at NIH among some of the best researchers in the world across so many fields. I see the groundbreaking work we are doing in projects such as ChIRP, BRICS, BTRIS, our deployment of a centralized REDCap instance, as well as new pilots for our Clinical Data Warehouse and self-service LLM-driven analytics tools and can’t help but be excited”, he says.
A record of success in the development of AI-driven clinical research analytics
Dobbins began his career as a research software engineer in UW Medicine at the University of Washington. Working within UW’s central data warehouse, he quickly learned of the challenges with integrating a dizzying variety of health data sources spanning multiple electronic health records, clinical service lines, ancillary systems, legacy databases, and more.
Beyond the technical difficulties of integrating such data, however, Dobbins was fascinated by the question of how to better enable researchers to safely and securely access and use data: “I’ve always believed strongly that researchers should not need to be experts in databases or computer programming languages in order to simply ask meaningful research questions. The driving question to my mind has always been, ‘How can I accelerate the research efforts of my colleagues and make accessing and leveraging data easier?’”.
Dobbins created Leaf, a friendly and intuitive self-service cohort discovery and data extraction platform capable of working with virtually any biomedical database, as well as running “networked” queries across multiple research organizations. Initially developed and piloted at the University of Washington, Leaf was open-sourced and is currently one of the most widely used cohort discovery tools at academic medical centers and research institutes around the world.
During his PhD, Dobbins studied under Professor Meliha Yetisgen in the renowned BioNLP lab at UW. In the days before ChatGPT, he also became an expert in training, fine- tuning, and evaluating LLMs to generate database queries for finding patients meeting clinical trial eligibility criteria. His work was the first published research to empirically demonstrate that machine learning models can surpass human experts in identifying eligible patients using actual clinical trials and structured database queries on real-world data. Building on this research, he developed LeafAI, a friendly self-service chat-like web-application which enabled researchers to enter free-text eligibility criteria and immediately view patients found in real-time.
He continued this line of research at Johns Hopkins and later Medeloop, where he explored the use of agentic LLM software systems for replicating published research studies in an automated fashion, extracting data from free-text clinical notes, and data de-identification.
Building on the CRISPI vision for research at NIH
Dobbins envisions a future for NIH IRP informatics and AI- based scientific tools to accelerate research and remove barriers: “I want our intramural investigators to be able to quickly pose nearly any hypothesis and have an AI-copilot tell them where to possibly find data, how many potential participants may be available, and what relevant published studies exist. All through a one-stop-shop web interface, which in turn connects to other agentic AI systems to assist with analysis, collaboration, and so on.”
He continues, “There are many challenges around this, including determining how to empirically evaluate LLM- driven systems so we can trust them. But NIH has long been a world leader in clinical informatics, and we have such tremendous potential to innovate that I firmly believe we will be a leader in this too.”
His current research interests include mechanistic interpretability, essentially reverse-engineering neural networks to determine how they “think”, and AI-driven database query optimization.
When he’s not working, he enjoys spending time with his family, including 3 kids, reading history books, studying Japanese language, and writing and recording music.
This page was last updated on Tuesday, September 9, 2025