About

AI teams building for healthcare need people who know exactly where clinical systems break — because they’ve been the one responsible when they do. I’ve spent 16 years in that seat: a dual-certified psychiatric-mental health and family nurse practitioner, a former hospital Medical Ethics Committee chair, and the sole clinician and IT operator of my own integrated practice.

The clinical side. I diagnose, prescribe, and manage risk in the domains where AI failures cost the most: psychopharmacology, controlled-substance stewardship, opioid use disorder treatment, ketamine protocols for treatment-resistant depression, and suicide and crisis assessment. When a model mishandles a multi-turn crisis conversation or a benzodiazepine request, I recognize the failure because I’ve managed the real-world version of it thousands of times.

The technical side. I designed and operate my practice’s entire technology stack — self-hosted clinical infrastructure, telehealth, secure messaging, and ambient AI documentation in daily use with real patients. I don’t evaluate clinical AI from a conference room. I deploy it, stress it, and live with the consequences in a live practice with my name on the door.

The governance side. I chaired a hospital Medical Ethics Committee and have served as an independent expert witness on standard of care and NP scope of practice since 2016. A decade of that work means I can translate between what clinicians actually do, what regulators and courts expect, and what an evaluation rubric or safety policy has to say for engineers to implement it.

What I do for AI teams

• Evaluation with clinical ground truth — model-output review and grading, hallucination and error detection, rubric and benchmark design for clinical and behavioral-health tasks. Under NDA, contributed to an ethics framework governing LLM behavior in a patient-facing clinical inference engine, translating autonomy, consent, capacity, and non-maleficence into evaluation criteria and model-output constraints; the framework was adopted into a widely used product and reduced escalation failures in high-risk clinical interactions.
• Red-teaming the hard scenarios — multi-turn crisis conversations, controlled-substance requests, capacity and consent edge cases, escalation and human-in-the-loop design.
• Governance that ships — turning clinical and ethical reality (autonomy, consent, PHI, documentation integrity, risk management) into evaluation criteria, safety protocols, and policy that engineering teams can actually build against.

Current practice

Founder and lead provider at North Branch Health, an integrated outpatient practice in Montpelier, Vermont — primary care, psychiatric medication management, treatment-resistant depression care, telehealth, and secure messaging. I intend to keep a clinical foothold as I take on new work — an evaluator who stops seeing patients starts grading against memory.

Background

Chair, Medical Ethics Committee, Gifford Medical Center (2016–2018). OAT/MAT addiction medicine at Treatment Associates. Rural family medicine across Vermont and Maine. Post-Master’s Certificate, Psychiatric Mental Health NP, Regis College; MSN, Family NP, University of Southern Maine; BS in organic chemistry and BA in anthropology, Evergreen State College.

I take on hard problems in clinical AI safety, evaluation, and governance — full-time or by engagement — work adjacent to clinical care, informed by every day I’ve spent inside it.