Healthcare AI product leader and founder.

RNA Research

My foundation in complex systems began at Georgia Tech, modeling RNA folding dynamics in Roger Wartell's lab. I am first author on a chapter in the ACS Symposium Series on the thermodynamics of sRNA–mRNA interactions and the role of Hfq, written as an undergraduate, and presented the work at FASEB and BMES and at the ACC Meeting of the Minds as one of eight students selected institute-wide.

Modeling a free-energy barrier is a search for the step that decides whether a process completes or stalls — the same question I now ask of systems that carry clinical consequence.

Epic

I had an MD admission in hand and deferred it a year to find out what healthcare technology was actually like. I joined Epic in Technical Services, owning what happened to eight client organizations after their software was installed — two on outpatient, six on the patient web portal. One 477-bed health organization had accumulated an outpatient issue list long enough to be a cancellation risk; working it to root cause turned it into a reference account.

I deployed MyChart and EpicCare Link for a 514-bed research hospital, taking a 134-year-old health system online for the first time, and wrote Caché plug-ins where the base product couldn't reach a department's actual workflow. I co-led CMS PQRS and built its escalation path — the direct ancestor of every escalation design I've built since.

Four months in I was recruited to internal area leadership; by the end of that year I was on the smallest team in a 6,000-person company, building the patient-facing portal — and I turned the medical school acceptance down.

The constraint was almost always the workflow around the clinician, not the clinician.

Doximity Dialer

Before Dialer, I authored Doximity's HIPAA compliance documentation and ran the largest national physician compensation survey to date — 35,000+ unique, ID-verified respondents — building the identity-verification infrastructure that made Doximity Dialer possible.

As founding product lead, I proved over a weekend that outbound calls could carry an unverified caller ID, then chose telco attestation over spoofing to clear institutional security reviews, built the compliance framework for hospital IT approval, and drove the Epic Haiku integration.

Across four years I built an enterprise marketing platform used by 6 of the top 10 U.S. research hospitals, founded the search team, and brought nurse practitioners and physician assistants onto the platform — 350K people who had not been able to join. I left as Head of Mobile Products & Growth, running every mobile app and user growth.

Dialer now carries more than 300,000 calls on an average workday across 250+ hospitals and health systems — #1 Telehealth Video Conferencing Platform in the 2026 Best in KLAS Report, its fifth consecutive year.

Trust is a product decision: what a system can prove about itself determines whether anyone acts on it.

NYU Stern

Doximity taught me that a technically sound product still has to clear an institution's own defenses — the compliance policy, the security review, the procurement process — before it can matter at scale. I took an MBA at NYU Stern to work that observation properly: not management as an end in itself, but finance, capital allocation, and organizational power as variables I'd been building around rather than reasoning with.

I tested the question on other people's companies before testing it on my own — evaluating seed-stage healthcare and deep-tech teams for General Catalyst's Rough Draft Ventures program, diligencing seed and Series A deals as an MBA Associate at the three-person, healthcare-focused Red Sea Ventures, and consulting to early-stage startups through the InSITE fellowship.

The pattern held everywhere I looked: the constraint on a good product was rarely the product. Andwise, three years later, was where I found out what happens when you have the authority to act on that observation yourself.

CancerCompass

At CancerCompass / CTCA Marketplace, I was the first product and technical hire for a pre-CEO, pre-revenue venture, and stayed through its growth from 2 to 25 people. I led a cross-functional team of 14 across three products — the oncology navigation platform serving 30MM annual visitors, a hospital digital-marketing platform, and a patient-engagement web app (CTCA was later acquired by City of Hope).

I reduced bounce rate by 25% and increased chat conversions by 267%, translating clinical protocol into steps a frightened family can act on at scale.

That's where I stopped believing information was the bottleneck: families in crisis didn't need more content, they needed navigation — options, sequencing, what to do tonight.

Transcarent

I joined Transcarent as an early product hire and stayed from the first activated member through 200+ employees; the company reached a $1.6B implied valuation over that period. I directed product across four value-based specialty-care and navigation programs (Surgery, everyday urgent care, Behavioral Health, and Oncology Care).

I led the Cancer & Decision Support pod — shipping CancerCare v1, a symptom-checking triage tool that inferred the right care pathway from reported symptoms, and an in-app second-opinion experience with automated handoff to Nurse Ally through Consumer Medical. I replaced manual case-by-case coordination with automated routing rules tied to clinical roles — the system assigned ownership of each case, and nurses and navigators shared the same facts during live care.

More coordinators was never the fix; explicit ownership was.

Andwise

As co-founder and CEO of Andwise, I raised $240K in initial funding, grew to 1,200+ physician users and a 700-member community, and convened a physician medical advisory board. I built the Contract Analyzer — automated clause-level analysis of physician contracts that flagged key terms for review — and review routing with escalation clocks so each analysis reached an accountable reviewer on a deadline.

Andwise proved more about the user problem than about the business: physicians used it, but the fundable paths would have made employers or financial institutions the economic customer, and the product would come to optimize for whoever paid. In 2024 I shut the company down rather than build that company.

The assumption I got wrong wasn't the product — it was believing enough user value would eventually produce a business model that preserved whose interests the company served.

Current Work

Today I lead AI product work for healthcare — the institutions, the workflows, and the trust gaps that decide whether AI actually runs. Across the healthcare products and the current AI work, I keep encountering the same class of problem: the important behavior lives outside the model — what the system can reach, what it is allowed to do, what its decisions depend on, how errors propagate, and whether people can challenge or correct them.

My current projects examine different parts of that problem. NextConsensus forecasts when medical evidence will force a guideline, FDA, or payer decision to change, and registers each forecast before the outcome is known so the track record can be checked rather than described. Ambit keeps what an agent system can do separate from what it may do. Refract watches the sources underneath a claim and flags what may have gone stale when one of them changes.

Fast Harm, Slow Repair measures the part nobody instruments: what happens after a system is wrong — how far the error travels and whether the correction reaches everything it touched.

I write about the wider pattern at The Crumple Zone and maintain Ethotechnics, an open framework for AI decision accountability.

The domain arrived at the questions I'd been accumulating: how systems earn trust, who owns their decisions, and what happens when they're wrong.

See also

• Work Case studies in healthcare AI product systems.
• Research The research program: what makes complex systems inhabitable.
• Résumé Complete timeline, education, and focus areas.
• Contact Email is fastest.