NOVA · HEALTHCARE / AI
Not every voicemail can wait.
An AI communication hub that helps oncology nurses see which patient voicemails need attention first.
NOVA turns incoming patient voicemails into one prioritized queue. Each call is transcribed, summarized and tagged by urgency, so the care team can scan, assign and act without listening to every message first.
I designed the product end to end, from the urgency system and case detail to the analytics view for managers, then built the prototype into a working React front end running on live data.
Role
Product Designer
Scope
End-to-end MVP
Tools
Figma, Figma Make, React
Industry
Healthcare · AI oncology
Timeline
2025
Problem
18,000+
patient calls.
One difficult question:
What needs attention first?
Every one of those calls landed as a voicemail. Symptom updates, follow-ups and appointment changes all arrived the same way, and a nurse had to listen to each one before anyone knew how urgent it was.
Before NOVA
Patient calls
Voicemail
Listen manually
Interpret
Prioritize
Route
Follow up
Dashed steps happened by hand, for every call, before urgency was even known.
The problem wasn’t listening to voicemail. It was knowing what mattered first.
Product Opportunity
From an inbox to a decision system.
Instead of a pile of recordings, the team opens a queue that has already been transcribed, summarized and sorted. The first decision is about the patient, not about which message to play.
01
Voicemail arrives
02
Transcribed
03
Summarized
04
Symptoms detected
05
Urgency scored
06
Queued by urgency
07
Assigned
08
Acted on & resolved
AI organizes the signal. The clinician still makes the decision.
Who I designed for
Two people, two questions.
Oncology nurses
“Who do I call back first?”
They work inside the queue all day. They need urgency, context and ownership at a glance, without opening every case.
Care team managers
“Are we keeping up?”
They step back from single cases to see response times, SLA compliance and where work is getting stuck.
Key decisions
Seven trade-offs, and why I made them
Each decision started from the familiar pattern. These are the places I chose differently, and the reasoning behind each one.
01
Instead of a dense data grid
Separated case rows
A grid fits more on screen, but every line looks equally important. Separate rows give each caller a clear boundary, so nurses scan case by case, not cell by cell.
02
Instead of urgency as one more column
Urgency first, at the far-left edge
Eyes start on the left. An urgency icon and a colored edge there mean priority is read before the caller’s name, not hunted for in a column.
03
Instead of loud red alerts
A subtle red SLA clock
When everything flashes, nothing stands out. A small red clock marks cases past their response window, and the Overdue count keeps the total in view. Visible enough to notice, quiet enough to avoid alarm fatigue.
04
Instead of an assignment modal
Inline assignment, in the row
Assigning used to mean opening the case, finding the field, choosing a name and saving. Ownership is the most frequent action, so it now lives in every row.
4 steps → 1
05
Instead of a separate case page
A slide-in case panel
A new page breaks a nurse’s place in the queue. The panel slides over the right side, so context opens without losing the list, and the next case stays in view.
06
Instead of one dashboard for everyone
An operational view and a manager view
Nurses ask “What should I act on next?” Managers ask “How well is the system performing?” One screen would answer both questions badly, so each role gets its own view.
07
Instead of a full EHR
Focus on communication triage
NOVA doesn’t try to replace the patient record. It focuses on the part that was failing: getting from a voicemail to the right person, fast, with the context intact.
01
Receive
02
Understand
03
Prioritize
04
Assign
05
Act
Nurse workflow
One queue, from voicemail to owner
The dashboard is where nurses spend their shift. Every step keeps the queue in view, so attention never has to start over.
01 · Scan
Urgency enters the scan path first.
Response windows by urgency
02 · Filter and assign
Narrow the queue, then take ownership without leaving it.
Case detail
Go deeper without losing the queue.
Opening a case slides a detail panel over the right side of the queue: caller, AI score, assignment, audio, summary, symptoms and history. The list stays right there behind it, and the panel expands to full width when a case needs more room.
Symptoms, clinical notes and call history, one scroll down.
Expanded to full width when a case needs the whole screen.
AI + human
AI can suggest. Nurses still decide.
NOVA’s AI does the reading: transcript, summary, symptoms and an urgency score. Every output stays visible, traceable and editable, so the nurse stays accountable for the decision.
01
An AI Score, shown as a number rather than a verdict
02
Every AI summary sits next to the full transcript
03
The original audio is always one click away
04
Priority, status and assignee stay editable
05
Call history logs what NOVA did automatically
Manager analytics
From individual cases to system health
The nurse dashboard answers “What needs attention right now?” Analytics answers a slower question for managers: “Where is the system starting to fail?” So I built it around gaps and trends, not totals.
01
Every number has a target
SLA compliance, cases pending over 48 hours and AI triage accuracy each sit beside their goal, so a manager reads the gap, not just the value.
02
Response time, split by urgency
Current vs target for Emergent, Urgent, Semi-Urgent and Routine. A healthy average can’t hide a slow emergent queue.
03
Volume next to workload
Communication volume by urgency sits with communications per staff, so rising demand reads as pressure on the team, not just a busier chart.
04
Unresolved cases as the early signal
Cases open longer than 48 hours are tracked on their own. A growing backlog is usually where a system starts to fail first.
05
AI accuracy, paired with overrides
Triage accuracy is shown alongside how often nurses override the AI. Trust in the model is something to measure, not assume.
Prototype screen. Figures are sample data used to design the structure, not measured results.
Scope · MVP
Designing the MVP also meant deciding what not to build.
Every feature had to earn its place in the core loop. If it didn’t help a nurse move a call from received to resolved, it waited.
NOW
Shipped in the MVP
Urgency-sorted voicemail queue
AI transcript and summary per call
Inline assignment and status
Case detail side panel
Filters, date and search
Manager analytics
OUT
Out of scope by design
Full EHR and deep patient context
NOVA is a triage layer, not a patient record. Leaving the EHR out kept the MVP on the loop that was failing (decision 07).
Visual language
Calm under pressure.
Eggplant carries the brand without competing with urgency. Neutral surfaces stay easy on the eyes through a long shift, so color is saved for what needs attention.
Eggplant
#45214A · Brand
Surface
#F5F5F7 · Canvas
White
#FFFFFF · Cards
Emergent red
#D93A3A · Urgency
Urgent orange
#EE8A3C · Urgency
Routine purple
#B79CD3 · Urgency
Work Sans
Headings and labels
Aa
Case Details · Call Insights · Dashboard
Inter
Data and body text
Aa
7 mins ago (Nov 4 · 10:32 AM)
Component details
Urgency reads as a label, a color and an icon, never color alone.
Summary and symptom tags sit in the row, so the scan doesn’t need a click.
The AI score is a supporting signal, sized smaller than the urgency it explains.
Figma to working product
Figma make helped this vision come true in a day
The design system moved straight into a React and TypeScript front end, wired to FastAPI hooks, so the queue ran on live product data instead of mock screens.
01
Figma
Screens and flows
02
Design System
Tokens and patterns
03
React/TypeScript
Front end
04
FastAPI
Data hooks
05
Live Product Data
The real queue
Product outcomes
What the design changed
Outcomes of the prototype and the working build. They describe what changed in the product, not measured clinical results.
Faster assignment, clearer priorities, and context that survives the handoff.
Faster assignment
Ownership is set from the row in one step, instead of four inside a modal.
Clearer prioritization
Urgency is the first thing read in every row, so the next call to return is obvious.
Better context preservation
Transcript, AI summary, audio and history open beside the queue, not instead of it.
Operational visibility
Managers see SLA compliance, response times, backlog and workload against their targets.
Human oversight of AI
AI scores and summaries stay suggestions. Nurses can check the source and change every field.
Working product foundation
React + TypeScript, on a live FastAPI backend
Not just screens: the frontend runs against a live backend, so the design had to hold up with real data, loading and errors.
Live cases
Statistics
Analytics
Optimistic UI
What I delivered
From workflow design to a working frontend
DESIGNED
End-to-end nurse communication workflow
Dashboard and triage
Case-detail interaction
Analytics
Authentication
Responsive component system
BUILT
Figma prototype
React + TypeScript frontend
Tailwind component system
FastAPI integration
Loading and error states
Optimistic interactions
Authentication: email or Google sign-in.
Reflection
What this project changed for me
I used to measure an AI feature by how much it could do. NOVA made me measure it by how clearly a nurse could see what it did, question it, and overrule it in seconds.
Good AI UX isn’t about making the AI feel powerful.
It’s about making the human feel informed and in control.
NOVA Health