Notes from the field.

The text-the-students problem.

The brief from the customer was a single sentence: "What if we texted students reminders to study?" That was all of it.

So I asked. Turned out nobody could say which subjects the students were actually struggling with. The outreach was generic. The admin team was hand-curating lists in spreadsheets. Nothing scaled, and nothing said whether any of it was working.

"What if we texted students" turned out to mean: "we have no idea who needs help with what."

I built a platform that identified each student's weakest area from their performance data, sent them a personalized link every morning, and gave admins a dashboard to see what was actually moving. The dashboard mattered as much as the texts. Without it, nobody would have known.

The photocopy upgrade.

Physical therapists were sending patients home with photocopied sheets of stick figures and hoping the patients did the exercises correctly. They were not. Nobody had any way of finding out.

I built Morphose: a HIPAA-compliant platform with 850+ interactive 3D exercise animations, progress tracking, secure messaging, and a therapist dashboard. The clinicians needed something they trusted in a regulated environment. The patients needed something that didn't feel like homework.

What I learned that applies to almost everything I've done since: when the user is a domain expert and the data is sensitive, you cannot ship and hope. You ship, then you sit in their office and watch them try to use it. The thing they don't say out loud is the thing you have to fix.

From a sensor on someone's arm to a chart on a screen.

Continuous glucose monitors are, on paper, an obvious product. In practice, the journey from "device measures glucose" to "doctor makes a decision based on a chart" is long and crowded with interesting failure modes. The data has to travel from BLE to phone to cloud to a clinician dashboard, and at every step it has to still mean what it meant when it started.

I built the visualization layer that doctors actually looked at: trends, time-series, the clinical context required to make a treatment call. The interesting work was almost entirely in the middle, where the data was bouncing between systems and had to be trustworthy on arrival.