Clinical ECG
Review Platform

Context

The client is a digital-health company behind a wearable cardiac monitor worn by patients over multi-day periods to capture continuous ECG data. The volume of data each patch produces is far beyond what any clinician could manually review, so the workflow depends on AI that pre-classifies and pre-flags potential arrhythmias. ECG technicians then use an internal review tool to check those AI outputs, accept or correct them, and finalize the report a cardiologist will sign.

The technicians are the human in the loop. The accuracy of the final report depends entirely on the speed and the quality of their review. When IBM Consulting was brought in, the brief was technically about cognitive load and review time. The underlying problem was harder: how do you redesign a workflow that practitioners trust, in a regulated clinical environment, without disrupting the muscle memory they've built up over years?

Approach

I led a team of four designers and researchers, owned the engagement vision in close partnership with the client-side product owner, alongside IBM engineering and clinical SMEs.

Discovery under access constraint. Practicing ECG technicians are not a population you can casually book for research sessions. They have patient backlogs, they work shift schedules, and access has to be negotiated through clinical leadership. We had limited windows with real users, which forced a different research approach.

The AI proxy. When direct technician access was unavailable, I built what we called a virtual AI ECG technician, a calibrated AI persona at multiple skill levels (tiered technicians, coaches, and scholars) that the design and product teams could use for early concept testing and edge-case exploration. It wasn't a replacement for real research; it was a pre-research instrument that let us catch obvious problems and arrive at the limited real-user sessions with sharper questions. Using AI to design for an AI-assisted workflow turned out to be a methodological win I'm still developing.

Outcome

• Improved throughput without sacrificing accuracy, validated against existing benchmarks.
• Strong qualitative response from technicians on the quality-of-life improvements. Quote-on-the-record feedback came from clinicians whose daily work was directly affected.
• A reusable AI-proxy research method that IBM teams have continued to apply on subsequent engagements where direct user access is constrained.

My Role

Design Lead. End-to-end ownership of discovery, research design, information architecture, interaction design, validation, and stakeholder communication. Led a team of four designers/researchers across the IBM pod and partnered directly with the client-side product owner throughout the engagement, with ongoing collaboration with IBM engineering and clinical SMEs.

What I'm Taking Forward

This engagement clarified what human in the loop actually means. It isn't a tagline. It's a specific design problem: a person looks at what an algorithm produced and has to decide what to do with it. The algorithm contributed something real. The person owns the call. Designing that moment is the work, and it's going to define product design for the next decade. I want to keep doing it.