Pippa

I embedded with clinicians, then shipped the platform into the clinics. Their feedback was the next release.

Pippa is the tool an outpatient eating-disorder clinic uses between visits, and the app the patient actually opens. I was in the conversations and in the codebase. 229+ clinicians described the job. I turned that into a roster, a chart, consent, and a patient app, and it went to production across 200+ clinics.

Problem

The customer could describe the fear and not the system. Patients would not use a logger that graded them. Clinicians needed intake, meal timing, and notes. A demo that showed calories to everyone would have failed the first clinic. The deployment had to land in a HIPAA environment, stay in sync when the patient was offline, and change when the next clinician call contradicted the backlog.

Process

I treated the calls as the spec. Repeated jobs became the chart: who logged, the meal, a note, a measure, a safety plan. The calorie model still runs on the photo, because the clinic asked for it, and the patient surface never draws the number, because they also asked for that. Daily check-ins sync patient logs to the dashboard. CI gates the release so a clinic is not the test environment.

Solution

• A production deployment, not a pilot deck: patient app and clinician dashboard, 10K+ daily users.
• An OpenAI companion and a vision model inside the clinical rule. The number exists. The patient does not see it.
• DynamoDB and automated check-ins so the chart and the phone agree in real time.
• A release path. Feedback from the clinics becomes the next ship, behind tests.

Results

• 229+ Clinicians in the room before the backlog was real
• 200+ Clinics in production
• 10K+ Daily active users
• 0 Calorie figures on the patient surface

Justin Moran · Forward Deployed Engineer · Chicago, IL · LinkedIn · GitHub · Resume (PDF)