PMcardio
01 Context 02 Outcome 03 Stages 04 Design 05 Impact 06 Learnings
HEALTHCARE UX · HEURISTIC EVALUATION · INTERFACE REDESIGN
Rebuilding Interface
of PMcardio
PMcardio interprets a 12-lead ECG from a single photograph. The model is validated. The experience around it was not.
I audited the existing app against usability heuristics, then rebuilt it.
ROLE
UX + UI, end to end
DOMAIN
Clinical AI · MedTech
METHOD
Heuristic evaluation
OUTPUT
Screens · Design System
01 CONTEXT
A validated model, wrapped in a flow that got in its way.
The diagnostic engine is backed by 15+ studies across 40,000+ patients. But reaching a reading took roughly fifteen screens, and the result asked clinicians to trust a number they could not interrogate.
THE AUDIT
I mapped the existing app across five flows that are pitch, enrolment, home, capture, and result. Approximately 33 screens end to end. Most of the distance to a first reading was friction: a countdown paywall, an unsourced accuracy claim, and consent buried in fine print.
METHOD — HEURISTIC EVALUATION
KEY STAGES
Audit, principles, restructure, build.
01 EVALUATE
Heuristic teardown
Audited approximately 33 screens across five flows, tagging every issue by severity to establish an order of operations.
02 DEFINE
Four principles
Time to value, Layered density, visible uncertainty. Every later decision traces to one.
03 RESTRUCTURE
Collapse the flow
Cut the path to a reading to four screens and rebuilt the result around plain language summaries, caveats, and per-lead confidence.
FINAL DESIGN
The result leads with plain language, then lets you dig.
A plain-language interpretation sits above an explicit statement: this is decision support, not a diagnosis.
Model names and method are demoted to a footer — present for the clinician who wants them, out of the way for everyone else.
Below are all the screen. Rebuild to provide the best Value per click. Interact with prototype here.
05 IMPACT
Trust, earned the honest way.
The reframed flow reaches a trustworthy reading in a fraction of the steps, and gives a patient a version they can understand without diluting the clinical detail underneath. Making uncertainty visible is the load-bearing decision.
FOR THE CLINICIAN
A second read that shows its working — per-lead confidence rather than a single opaque score.
FOR THE PATIENT
Plain language first, with clinical depth available rather than imposed.
FOR THE PRODUCT
Credibility built on sourced evidence instead of manufactured urgency.
A clinical AI should never ask for trust it hasn’t earned.
PMcardio · UX + UI case study · 2026