Track record

Situation

Resmed shipped devices at the right clinical pressure, but the comfort settings that keep patients on therapy long term went untuned. A third of patients quit within a year, each one a multi-thousand-dollar loss in lifetime revenue.

What I did

I made the case to the executive team, secured funding, and set the technical direction: digital twins, pairs of near-identical patients in the historical data, one on tuned settings and one untuned, producing real-world evidence that comfort tuning causes adherence. That evidence cleared FDA review. I ran regulatory, clinical, and commercial in parallel with engineering so the submission and the product moved together instead of in sequence.

Situation

The condition is under-diagnosed at enormous scale, close to a billion people worldwide, most of whom never connect their symptoms to a treatable problem.

What I did

I bet on general-purpose LLMs early, before most enterprises would put them in production. The hard problem was clinical trust: an assistant that guides a stranger toward diagnosis without ever giving medical advice. I grounded it with retrieval over a vetted clinical corpus and brought clinical, legal, regulatory, and marketing along with engineering so it could actually ship.

Situation

Partners are reimbursed only when a patient clears a 90-day compliance threshold, and a quarter miss it. A small front office cannot chase a million active patients.

What I did

I directed the build of machine-learning models that scored each patient's daily risk of missing compliance from device telemetry and clinical data, turning the score into a prioritized daily worklist: who to call, when, and what to try. A rules layer built from partner feedback fit it to their workflow rather than adding work.

Situation

A pandemic supply disruption and a competitor recall hit at once, sending demand surging. Freight ran north of $150M a year, with air-versus-sea calls made on gut feel.

What I did

This was supply chain's territory, not mine, so I earned the right to put an algorithm into it by bringing operations, finance, and IT in from the start. I set the direction, mixed-integer optimization with ML demand forecasting, to produce the optimal air and sea split per shipment, after first unifying the freight, ERP, and demand data into one foundation the model could run on.

Situation

The company sat on more than 8 billion nights of therapy data, one of the richest datasets in healthcare, and almost no ability to use it. No vision, no platform, no operating model.

What I did

I was the primary architect of the AI vision, won executive buy-in and a multi-million-dollar annual budget before there was a track record, and built a team of roughly 60 specialists, about 20 of them PhDs, across four geographies chosen for talent and cost. I set the operating model that moved work from data-science insight to shipped, regulated product. Most AI organizations never build that bridge.

Situation

In a business where regulatory trust is non-negotiable, AI was scaling across regulated and consumer products on the same data, and the existing quality system was not built for it.

What I did

I redesigned the quality management system for AI and directed the build of a patented AIOps platform that baked compliance into the pipeline by design: model monitoring, audit trails, lineage, drift detection, and automated deployment, across dozens of models running tens of millions of daily inferences. I chaired the AI product committee and led AI fluency education for the board and C-suite.