- The problem
- Nurses on the pilot ward had learned to silence the monitor before they had finished reading it. Nine alarms in ten were false, and the one that mattered arrived in exactly the same red as the nine that did not.
- What was made
- A confidence model pushed all the way to the surface. Every reading carries the sensor’s own certainty, the display degrades honestly when a lead is loose, and a triage board ranks patients by how much the system trusts what it is seeing rather than by raw threshold.
- The outcome
- Alarm volume down 61% across the pilot ward. Median time to escalation for genuine deterioration fell from nine minutes to four. Cleared class IIa review with no interface findings raised.
RoleLead product designer — research, design system, front-end prototypes
Year2026 · 14 months · with a team of five