I build clinical AI and NLP systems ✨ at Boston Children's Hospital 🏥 on the @smart-on-fhir / Cumulus team ☁️. Over the past ~9 years I've contributed to numerous full-stack web development projects 🕸️ in the Healthcare interoperability space; in the past ~2 years I've been working on LLM-based clinical information extraction, designing pipelines that turn unstructured clinical notes into structured, research-ready data across multiple disease studies. I earned my CS Master's from Tufts in 2021 🐘, focused on AI/ML and AI ethics 🤖⚖️ for public-sector and healthcare applications.
🔬 What I'm working on now
- Clinical LLM extraction at scale. I work in the Cumulus ecosystem to enable a model-agnostic NLP extraction platform that runs frontier, open-weight, and locally-hosted models against real FHIR clinical-note corpora, with MLflow experiment tracking for token economics, model parameters, and extraction metrics.
- We build Cumulus-ETL and Cumulus-Library to extract population-scale data across partner health-sites.
- We also support a number of specific studies across disease-specific study libraries to enable exploration of domain-specific research questions. Some areas we're using cumulus and LLMs to explore include the relationship between pediatric IBD outcomes and medications, immune-related adverse events in kidney-transplant patients, and more!
- Evaluation & annotation. I translate clinical chart-review guidelines into Label Studio annotation interfaces so clinicians can produce the gold-standard datasets our extraction pipelines are measured against.
- Research & publishing. I work with my colleagues to publish on the impact that interoperability and LLM systems can have on multiple levels of clinical care.
🔭 Previously
- @MITRE: advancing oncology through data standards like FHIR and mCODE — we built the mCODE Extraction Framework to enable structured data extraction in support of an oncology clinical trial, and we also built FluxNotes to imagine what the future of multi-modal data capture could look like – before LLMs 😆!
- 🌱 Ongoing interest in fairness, accountability, transparency, and power in healthcare algorithms — and what technologists, regulators, and citizens can do about them.
- 👯 Local civic-data projects like visualizing Arlington Town Meeting Member voting histories with @GrahamGoudeau and @mgramigna.
📫 Reach me: email preferred — dtphelan1<<at>>gmail<<dot>>com · 🌐 dylanphelan.tech · 😄 Pronouns: he/him