Duke Clinical Research Institute

Dylan Thibault

Senior Biostatistician building clinical AI systems that hold up in real workflows.

I work at the intersection of biostatistics, healthcare data, and AI product delivery, translating rigorous research practice into tools teams can actually adopt.

115+

publications

10+

years clinical analytics

2024

DCRI Innovator Award

About

Clinical research meets product execution

I'm a Senior Biostatistician at the Duke Clinical Research Institute, where I lead statistical work and AI initiatives across registries, trials, outcomes research, and internal tooling.

My background spans Medicare data, risk prediction, propensity methods, dashboard development, and collaborative publication work across more than 100 papers. The more recent layer of my work is building AI products that make statistical and clinical workflows faster without lowering the bar.

Biostatistics leadership

Cardiovascular registries, clinical trials, health outcomes research, and modeling work that has to stand up to scrutiny.

AI system design

Workflow automation, retrieval systems, QA agents, and domain-specific interfaces for research and clinical operations.

Team enablement

Training analysts and researchers in AI-assisted workflows, modern coding practices, and practical tool adoption.

Education

MS Biostatistics — Georgetown University

BS Biological Sciences — University of Vermont

BA Philosophy — University of Vermont

Recognition

DCRI Innovator Award recipient in 2024 for ChatSum, a meeting-minutes AI product built to reduce coordination overhead for research teams.

Consulting

I help organizations use AI with a bias toward working systems, clear operating constraints, and measurable usefulness. The best engagements sit at the boundary between domain expertise and implementation.

01

AI strategy and workflow design

Find the highest-leverage places to introduce AI into research, operations, and analytic workflows without creating governance problems later.

02

Custom internal tools

Build focused applications for document processing, review, consent, summarization, and other domain-specific workflows.

03

Agent and RAG systems

Design retrieval pipelines, QA agents, and human-in-the-loop review patterns that fit regulated or research-heavy environments.

04

Team training

Teach teams how to use LLMs, prompt systems, automation, and coding tools in ways that are practical, safe, and repeatable.

Best fit

Teams that need both domain judgment and execution.

Research groups, healthcare operators, and product teams tend to get the most value when the work requires both analytical depth and hands-on delivery.

Typical outcomes

Clear workflow map, scoped tool concept, implementation plan, or shipped prototype.

Open source

I also build with and advise on open-source tools like OpenClaw.

Start a Conversation

Selected AI Work

The strongest work is not generic AI experimentation. It is domain-specific systems that reduce friction for research teams, reviewers, and patients.

Award-winning internal product

ChatSum — Meeting Minutes AI

GPT-4o powered meeting minutes summarizer. Won the 2024 DCRI Innovator Award.

Why it matters

The best tools remove coordination overhead for clinicians, analysts, and project teams instead of adding a layer of AI theater.

Built with

GPT-4oProductivityAward-Winning

More systems

Automated SAP Generation

Next.js app using retrieval-augmented generation (RAG) to generate Statistical Analysis Plans. Upload a protocol, receive a tailored, downloadable SAP.

01
RAGGraph DatabaseNext.jsPython

Agent-Based SAP Review

CLI tool with dual-agent framework (drafting + reviewer agents) that iteratively refines SAPs against CRO/industry guidelines.

02
Multi-AgentCLIQuality ControlPython

AI Informed Consent Platform

Real-time patient-facing chatbot for informed consent evaluation. FastAPI + Next.js with GPT-4o/5, evaluating 17 quality metrics for regulatory review.

03
FastAPINext.jsGPT-4oReal-time

COVID Resource Calculator

Predictive calculator for hospital resource utilization during COVID — blood products, ventilation, LOS, readmission.

04
Risk PredictionCOVID-19Clinical Decision Support

Azure AI Studio

AI-powered content generation studio with TTS, video, and image capabilities using Azure OpenAI.

05
AzureOpenAITypeScript

Publications

60+peer-reviewed publications

Live search powered by the PubMed database.

Searching PubMed…

Experience

The through-line is long-form analytical work paired with increasing product, mentoring, and AI implementation responsibility.

Senior Biostatistician

Duke Clinical Research Institute (DCRI)

2021 – Present
  • Lead AI initiatives: automated SAP generation with RAG, agent-based SAP review, AI-enabled informed consent platforms
  • Statistical analysis across outcomes, registry, claims, and administrative datasets
  • Primary supervisor/mentor for junior biostatisticians
  • Department-wide AI training and biweekly AI office hours
  • DCRI Innovator Award recipient (2024) for ChatSum meeting minutes AI app

Biostatistician III

Duke Clinical Research Institute (DCRI)

2016 – 2021
  • Lead statistician on the Society of Thoracic Surgeons (STS) registry across all three datasets: adult, thoracic, and congenital
  • Published 50+ peer-reviewed manuscripts including risk prediction models
  • Built COVID hospital resource utilization predictive calculator

Biostatistician

University of Pennsylvania, Department of Neurology

2013 – 2016
  • Data analysis on NIS, Medicare CMS, KID, and NEDS datasets
  • Statistical methods: survival analysis, mixed effects models, propensity score methods
  • Published multiple first-author manuscripts in peer-reviewed journals

Projects

Some work is open and visible. A larger share lives behind institutional boundaries, but the technical patterns are consistent across both.

Private systems

informed-consent-aiPrivate

AI-powered chatbot to help patients understand clinical trial informed consent documents.

Python
gwtg-sap-appPrivate

Next.js app for GWTG statistical analysis plans using RAG.

Python
sap-generatorPrivate

CLI for creating statistical analysis plans with AI assistance.

Python
pre-emptPrivate

PRE-EMPT chatbot for clinical trial screening and evaluation.

Python
azure-ai-studioPrivate

AI-powered content generation studio with TTS, video, and image using Azure OpenAI.

TypeScript
teams-transcript-downloaderPrivate

Download and organize Microsoft Teams meeting transcripts via CLI or web interface.

Python
azure-capacity-checkerPrivate

Web dashboard to check Azure OpenAI model capacity across regions.

JavaScript
Daily76Private

This day in American history — daily trivia webapp.

TypeScript

Capabilities

Working range

The common thread is not any single tool. It is the ability to move between rigorous analysis, product thinking, and implementation without losing the shape of the problem.

Statistical Methods

Survival AnalysisRisk Prediction ModelingPropensity Score MethodsMixed Effects ModelsBayesian StatisticsModel Calibration & Validation

Programming

PythonRSASTypeScriptSQLGit/GitHub

AI & Tools

LLMsRAG / Vector SearchMulti-Agent SystemsFastAPINext.jsStreamlitAzure OpenAIClaude CodeCursorCodex

Domains

Clinical TrialsCardiovascular Registries (STS)Medicare/Claims DataHealth Outcomes ResearchInformed ConsentFDA Regulatory Pathways

Get in Touch

Let's build something useful.

If you're exploring a clinical AI workflow, an internal research tool, or a team enablement problem, send a note with enough context to make the first conversation concrete.

Consulting engagements and scoped AI tool work

Research collaboration, product strategy, and implementation planning

Speaking, team training, and workflow coaching

Start with a short note