Enter the password to view this case study.
I designed a scheduling system for hospital floors — the kind of place where a 45-minute phone scramble means a ward goes understaffed. No training manuals. No onboarding sessions. It just had to work.
That's what US hospitals waste annually on staffing coordination — agency nurses, overtime, manual scheduling. Nurse managers spend eight-plus hours a week juggling spreadsheets and phone trees. Floor staff use legacy terminals from the last decade and have zero patience for new software. I led a team of six to design a system so intuitive that nobody would need training to use it.
A nurse calls out sick. The charge nurse starts working down a phone list — calling, texting, leaving voicemails. Forty-five minutes later, maybe someone picks up. Maybe the ward goes short-staffed.
Every unit has its own scheduling spreadsheet. No single source of truth. No audit trail. Union rules, seniority, overtime thresholds — all tracked in someone's head or not tracked at all.
Schedule changes spread through word of mouth. Shift swaps require paper forms. PTO requests sit in a manila folder on someone's desk. Staff find out about changes after the fact — or not at all.
These constraints weren't obstacles — they were the design brief. Every decision I made was shaped by the reality of who uses this system and where.
Nurses' station terminals run outdated Chromium on 1366×768 screens with resistive touchscreens. I designed for the worst monitor in the building, not the best one in our office.
WCAG AA wasn't a nice-to-have — it was mandated. Every component hits 4.5:1 contrast. Full keyboard navigation. Screen-reader markup. No information encoded by color alone.
Seniority, mandatory rest periods, overtime thresholds, certification requirements — all of it has to be surfaced inline, at the moment of decision. Miss one rule and you trigger a grievance.
Every critical flow — accepting a shift, approving a swap, checking coverage — has to complete in under three taps and three seconds. That's the ceiling. No exceptions.
Before I opened Figma, I spent time on the floor. Shadowed charge nurses during shift changes. Watched managers wrestle with spreadsheets. The research shaped everything that came after.
Shadowed nurses across three facilities. Audited every competitor in the space. Interviewed managers, floor staff, and union reps to understand the real workflow — not the documented one.
Mapped the information architecture for two very different personas: managers who need density and overview, nurses who need speed and clarity. Defined subsystems and a constraint framework.
Four versions from wireframes to high-fidelity. Weekly design critiques with the team, bi-weekly stakeholder reviews. Each round killed features that looked good but didn't survive the floor.
Tested 13 manager workflows and 26 nurse-facing components in structured usability sessions. The PTO request flow failed twice before we got it right.
I designed the UX layer for an AI matching engine that does in thirty seconds what used to take forty-five minutes and a lot of voicemails. Three steps, fully automated, with the manager always in the loop.
The system monitors the schedule in real time. Call-out at 4 AM? Census spike in the ICU? It spots the gap before the charge nurse does and flags it immediately.
The AI ranks available staff by certification, union seniority, overtime cost, and rest-period compliance. The manager sees a scored list — not a phone book — and can override any recommendation.
One push notification to the best-matched nurse. One tap to accept. Shift filled, manager notified, schedule updated — all before the coffee gets cold.
Built the core loop: manager dashboard plus nurse mobile app. The question wasn't whether the technology worked — it was whether people on a hospital floor would actually use it instead of what they've done for twenty years.
Drawing on my BFA in art direction, I led the brand through seven logo iterations, testing how the mark communicates urgency, care, and reliability. Landed on a purple-dominant palette with a typographic system that works at 5px on phone screens and 24px on dashboards.
This was the turning point. Instead of managers calling down a list, we introduced the Shift Marketplace — open shifts posted like listings, nurses browsing and claiming. The phone tree became optional overnight.
Structured usability testing across 13 manager workflows and 26 nurse components. The tests confirmed the zero-learning-curve hypothesis — and surfaced a PTO request friction point we fixed before dev handoff.
Everything a charge nurse needs at 5 AM: pending approvals, unfilled shifts, coverage gaps
Daily schedule, shift offers, and real-time notifications — designed for one-handed use between rounds
Open shifts listed like a marketplace. AI scores candidates by certification, seniority, and compliance — managers see who fits, not a phone list
Most healthcare software looks clinical — sterile blues, stock photography of smiling nurses, forgettable. I wanted something that felt modern and trustworthy without defaulting to the same visual language everyone else uses.
Each exploration tested a different idea about what "urgent" and "reliable" look like as a symbol. Most of them taught me what not to do.
Encoding union seniority, overtime thresholds, and certification requirements into an interface that never makes you think about them — that's harder than it sounds. And it's the whole job.
Managers need data density. Nurses need speed. Serving both with one visual system meant constant negotiation — and a few decisions that felt wrong until they were tested.
Legacy hardware, accessibility mandates, union rules, and a three-second performance ceiling didn't limit the design. They defined it. Every hard constraint eliminated a bad idea.
The happy path always works. Swapping a shift thirty minutes before it starts, triggering an overtime grievance, recovering from a mid-flow browser crash — that's where the design actually matters.
A multi-agent AI marketing platform with a 62% drop-off rate. I slowed the system down on purpose — and that's what made it work.
Read the case study →