85%
of clinician-collected patient questions answered in testing — 17 of 20 — by the dashboard + chatbot
virtualfieldRedesigning how glaucoma is seen —
by the clinicians who track it, and the patients who live with it.
VirtualField's VR headset made visual field testing fast, portable, and accessible. I redesigned what happens after the test — turning its clinically rich reports into a longitudinal dashboard, a patient view, and an AI chatbot — a working prototype, accepted by VirtualField's CEO as a direction for future development.
3 assets
one connected system: clinician dashboard · patient view + handout · chatbot
6→1
successive reports collapsed into one longitudinal dashboard
~3 wks
discover 9d · define 3d · design 7d · deliver 1d
Everything in this clinic gets filed — watch for the patient folders as you scroll.
The redesign, live — zoom, filter, hover
virtualfield
MRDr. Martin Reiner
JD
JANE DOE62 · Female · P.ID: YG367T29ND9987
OphthalmologistMartin Reiner, M.D.
DiagnosisOpen-Angle Glaucoma
Treatment StartedApr 1, 2019 · 6 appts
Next AppointmentApr 1, 2022
Visual Field Index & Treatment Summary
VFI-L: 58%VFI-R: 74%
Scroll on the chart to zoom, or use the controls and date filter. Predictions cover both eyes — hover the endpoints for the with- and without-treatment numbers.
Act I The situation
Meet VirtualField.
The company
VirtualField (today rebranded as Carrot) makes a VR headset that tests peripheral vision — faster and more portable than the tabletop machines it replaces.
The test
A visual field test maps how much of your side vision remains. The score is the VFI — Visual Field Index — from 100% (a full field) down.
The disease
Glaucoma narrows that field so gradually patients don't notice.[1] It can't be cured; treatment slows it.[2,3] So patients are tested again and again, across years.
The report
Every test prints a four-page progression report — grayscale maps, threshold grids, a summary table. It's what the doctor reads, and what the patient sees.
VirtualField asked me to redesign this report.
Act I The problem
One patient.
Six appointments.
Six documents.
Now consider Jane Doe — a 62-year-old patient with open-angle glaucoma who has been coming in since 2019. Her entire history lives in six of these documents.
4 pages
per appointment, per patient review
2–3
appointments legible per printed chart before lines compress
×6
documents holding one patient's history
Rebuilt from a real VirtualField report. Hover a page.
Act I The research
Before redesigning anything, I asked the people who read these reports every day.
Research · Firsthand interviews
Every clinician worked through the same structured guide, and I listened for two things at once.
12 clinicians — ophthalmologists, optometrists, and clinical technicians who use the VirtualField headset in practice.
30-minute remote sessions with a structured moderator guide, flexible enough to probe friction moments.
Two lenses: their own workflow — and the patient questions they field in every appointment.
Synthesis: affinity mapping in FigJam, then an Excel tally of pattern frequency — the two problems below are the ones that kept scoring.
Patient voice, proxied: every patient insight in this case study arrives through clinicians — the strongest lens available under clinical access constraints, and the first limitation I’d address in a next round.
Finding 1
Fragmentation
The data was all there — spread across pages and sessions. Comparing visits meant holding the last chart in memory while reading the next, inside a fifteen-minute slot.
Finding 2
Confusion
Patients couldn't read the reports at all. Clinical terms, unfamiliar graphs, red numbers with no context — more alarming than informative.[4]
Finding 3
A gap nobody owned
Some patients stopped their drops, believing treatment was temporary. Others never came back. In the absence of understanding, they filled the gap with wrong assumptions.[5]
Act I The tension & the reframe
The ask was a better-looking report.
Research revealed two deeper problems.
01
Glaucoma is a story told across years.
A printed page shows one appointment. The redesign had to be scalable across time — as legible at appointment twelve as at appointment two.
02
The users had very different requirements.
The redesign had to accommodate mental models across levels of expertise — and the research showed how far apart they sit:
Clinicians
needed dense, comparative data.
Every value, every visit, side by side. Density is confidence.
Patients
grew more anxious as the data grew denser.
The numbers that reassure a doctor frighten the person they describe.
The paper report format wasn't built for the story it needed to tell.
How might we
…turn a static test report into the longitudinal story clinicians need to tell — and patients need to understand?
Every data point stayed clinical-grade.What changed was how it was presented — to two different audiences.
Here the case study turns — Act I diagnosed the paper report; what follows is the system that replaced it.
Act II From sketch to system
The messy middle, filed — seven steps to the system.
The first version was a stack of notebook sketches. From there: layouts, a grayscale wireframe, a color pass, a carousel concept, then the split into separate clinician and patient dashboards. Three of the seven steps sit below — click a wireframe or the final system to scroll through it. The full progression, including every intermediate version, lives in the patient folder.
Three of seven steps — the full progression lives in the folder below
Act II The system, in the light
01
Clinicians got the whole arc, in one view.
A longitudinal dashboard replaces the paper stack: one zoomable VFI chart across the full patient history, both eyes side by side, and an AI trajectory showing what happens with and without treatment — the concrete artifact for the hardest conversation clinicians described: "why keep coming back when I feel fine?"
What to look at: the fork after the last recorded appointment, where two futures open — flat with treatment, falling without. Clinicians told me the hardest part of a visit is making an invisible disease feel real. This is the picture they can point at: the left eye lost 22 points in under three years; the projection holds at 58 with the drops, and keeps falling without them.
Hover any point — the zoomable, date-filterable version is the chart at the top of this case study
Below the chart, nothing asked clinicians to relearn anything. The tables keep the printed report's exact formatting — the same column labels, the same red flags — because twelve interviews made one thing clear: don't cost a clinician a single second of processing. What's new is what paper couldn't do: sort it, filter it by eye, and ask any column what it means.
Hover a column heading for its definition · click to sort · switch eyes above
And research showed doctors read change by comparing successive appointments. So the charts pair consecutive visits side by side — every plot, the visit's key values, and the doctor's notes in one frame:
Step through Jane's six appointments — five successive comparisons per eye
02
Patients got the same data — asked to do a different job.
The patient dashboard opens with the same chart. That was deliberate: patients deserve to see their own clinical data. What changes is everything around it — three wins, mirroring the clinicians'. The screens below are the actual Figma build.
What 100% means. Why a flat line is the win. Why treatment matters even when nothing hurts. Written at the level of the questions clinicians said patients actually ask.
From the Figma build · patient dashboardUnderstanding your journey
Understanding is one thing; knowing what to do is another. The Healthy Sight checklist turns care into five concrete habits — each one small enough to actually keep.
From the Figma build · patient dashboardHealthy Sight checklist
The FAQ answers the questions patients carry home — honestly, including the hard one. Try it:
03
And a chatbot for the questions patients are too anxious to ask.
Not in the brief — the idea surfaced in a clinician interview, and the research kept pointing back to it: the dashboard answers what the data says, not what a patient needs to ask, alone, in her own words. I prototyped it as a Custom GPT — fast to iterate, no engineering dependency — and put it in front of clinicians twice.
Designed to do three things: explain plainly, navigate to where the answer lives on the dashboard, and escalate to the doctor at its edges.
Conversation design · from the chatbot's instructions
Answer only from Jane's own records — her VFI history, medications, appointments. No general medicine.
Two sentences, then point to where it lives on her dashboard. The bot is a guide, not a destination.
Plain words: “side vision,” never “peripheral field defect.” Reading level ≈ a good nurse, not a journal.
Never diagnose, never adjust treatment. Some questions are a call — the bot says so and points to the handout.
Usability testing · Two rounds, clinician proxies
Trust was the metric. Two rounds moved it.
Round 1 — 2 clinicians, remote Zoom sessions with screen-shared access to the live prototype.
The bar: how many attempts before they’d trust a response in front of a patient. Round 1’s answer: 2–5. Verbose, ungrounded answers eroded trust fastest.
Two fixes between rounds — the persona was rewritten with conciseness as an explicit trait, and Jane’s eye history became structured reference data.
Round 2 — 1 clinician, same criteria: responses came back precise, historically grounded, and trusted. Accepted by stakeholders as a direction for future development.
The decision I'd defend in any interview:
Nobody in severe pain logs into a portal and types.
Emergency guidance lives on the printed handout — not in the chatbot.
Round 1 shipped the bot with an emergency pathway: severe pain, sudden vision loss, call this number. In reflection I cut it and moved the guidance to the take-home report — already on the fridge, readable with no device, no login, no power. Context-aware design over feature-complete design.
From the Figma build · the printed handoutClick to zoom
Same questions, before and after the fixes
Round 1 · before
Round 2 · after
Representative exchanges, rebuilt from the prototype — Round 1 wording reconstructed from testing notes
Act II The journey
Same patient. Same diagnosis. A different journey.
Here is Jane's path through a testing cycle, before and after — the same five moments, a different artifact in her hands at each one, and a different feeling walking through it.
Click the map to expand it full-screen
Act II The design system
One design system. Two emotional registers.
Both dashboards share the same foundations — tokens, type scale, chart styles, table patterns, controls. What the system enables is calibration: identical components applied at different densities for different emotional contexts. Same tokens, different application.
Typography — live specimens
H1 · Syne 700 · 30/1.1Progression Report
Section · Syne 600 · 19 capsVisual Field Index
Body · Manrope 400 · 14.5Your condition has been stabilizing since April 2020.
Label · Manrope 700 · 11 capsTreatment summary
Color tokens
left eye
#054FB9 right eye
#F57600 left · 50%
prediction right · 50%
prediction text / dark
#202320 pale blue
info surfaces teal
highlighted flag red
cell highlight
#054FB9 right eye
#F57600 left · 50%
prediction right · 50%
prediction text / dark
#202320 pale blue
info surfaces teal
highlighted flag red
cell highlight
Buttons — hover them
Data patterns
| Date | VFI-L | VFI-R |
|---|---|---|
| Apr 1, 2021 | 62% | 75% |
| Oct 1, 2021 | 58% | 74% |
Left eyeRight eye
Rebuilt in code from the Figma library — every specimen above is live HTML and CSS, not an image.
Act II Accessibility
Designed for eyes that are losing sight.
The audience for this dashboard is, by definition, people with impaired vision — so accessibility wasn't a checklist at the end, it was the working constraint. The system was tested to hold up across eye disabilities, built on four principles:
Attention
- Minimal use of color — an accessible palette with strong contrast against the background
- Highs, lows, and trends legible at a glance
- Clear hierarchy in section headings and text
Perception
- The data holds its integrity in grayscale
- Left and right eye read in one shared context
- Complementary colors paired with distinct shapes — the blue diamond and the orange circle
Memory & cognitive load
- Tables follow the format clinicians already know — prior expertise does the work
- Plain-language summaries carry the story for patients
- Clear labeling, proportionate axes, and a single chart for both eyes
Interaction
- The data can be filtered — date ranges, eyes, zoom — so every reader controls the density they see
Reflection One honest thing
The goal was never to make patients experts in their disease — only supported enough to navigate it without being one.
Usability testing showed reduced confusion and increased confidence in reading results. Clinicians confirmed the patient content maps to the questions they actually field.
“You've taken it further than we could've imagined! The patient dashboard and handouts are great assets on their own. While the handouts are something we can start implementing right off the bat, I can genuinely see the dashboard becoming part of our development pipeline in the coming months!”
Why I'd choose it again
This is the work I'd choose again, every time. Healthcare, with real clinical stakes. Two user groups sitting poles apart on the same data. And genuinely end-to-end range: research, product design, usability testing — and conversational design, which turned out to be a craft of its own and my favorite part of the build.
Trade-offs I accepted — and why
- Clinicians stood in for patients. Access constraints; they field patient questions daily — but proxies have limits.
- Longitudinal data was simulated from sequential report structures — no live datasets existed to design against.
- Adherence went unmeasured. It needs months of follow-up; the project had three weeks.
With more time or resources
- One or two direct patient participants — to stress-test how patients feel, not just what they ask.
- A longitudinal study: do dashboard patients return for follow-ups at higher rates than printout patients?
- Test the handout as an artifact — does it measurably reduce misconceptions by the next visit?
References
[1] American Academy of Ophthalmology — What Is Glaucoma? Open-angle glaucoma shows no early warning signs; blind spots grow in peripheral vision and most patients notice nothing until damage is severe · aao.org
[2] Glaucoma Research Foundation — Symptoms of Glaucoma: vision loss from glaucoma is not reversible with treatment, even with surgery · glaucoma.org
[3] National Eye Institute — Glaucoma: sight is lost slowly from the periphery; without treatment glaucoma can eventually cause blindness · nei.nih.gov
[4] Ophthalmic & Physiological Optics (2023) — systematic review of adherence barriers: poor understanding of glaucoma is consistently reported by patients and practitioners as a driver of nonadherence · doi.org/10.1111/opo.13245
[5] Ophthalmology Glaucoma (2024) — Patient Challenges with Glaucoma Eye Drops: roughly half of patients discontinue their glaucoma medication within the first six months, adherence undermined by an initially symptomless disease · ophthalmologyglaucoma.org
Open to full-time roles
Reach out for a coffee chat.
I’m looking for full-time product design and UX research roles, and I’m happy to relocate.
If you have something that needs research behind it and someone who can take it all the way to a shipped interface, I’d like to hear about it. In person over actual coffee, or virtually — whichever is easier.