Clinical intelligence platform · Case study

Turning scattered medical records into clear clinical decisions.

N1 Care is an AI clinical intelligence platform for licensed healthcare professionals. I rebuilt the doctor-facing workspace around one practical question: what does the clinician need to understand, and what do they need to do next?

Product
N1 Care, by N1 Healthcare
Role
Design lead · sole designer
Timeline
Aug 2025 to Feb 2026
Users
Doctors and clinical staff
Team
~23 people, one designer
Scope
Audit, UX, UI, prototypes, testing
N1 Care Today worklist, ordered by clinical priority, with patient, concern, urgency and action aligned in columns
Fig. 01The home screen became a clinical worklist. Each row keeps the patient, the concern, the urgency and the next action on one line.

Rebuilt demo · placeholder product name · fictional patients · local-only data · no client service connected

01

What I inherited

The product had grown in Bubble.io before the team began moving it to React. It could already collect records, extract clinical data and generate reports. But the interface had accumulated the kind of debt that makes a capable product feel slow and difficult to trust.

The staging environment repeatedly timed out on everyday create, search and pagination actions. I cannot prove Bubble itself caused every delay, but the experience was slow enough to interrupt basic review work.

20+

Open UI/UX issues

The migration carried a visible backlog of layout, interaction and consistency problems.

3+

Interactions to one graph

There was no useful trend signal until a doctor opened each biomarker and changed tabs.

48

Button variants

Similar actions looked and behaved differently across the product.

86

Documented colours

The visual system had more options than the interface could use consistently.

Source: internal product audit and migration records, Jan–Feb 2026

The real problem was interaction debt

A complex patient can arrive with years of labs, medications, diagnoses, procedures and scanned reports. The information exists, but it is spread across formats and screens. The doctor still has to assemble the story while making a decision.

The interface had the same problem. Each module could display its own data, yet the full workflow never made priority, safety, chronology and evidence feel connected.

01 / Records

Evidence arrived in pieces

PDFs, images, lab results and historical notes each carried part of the clinical picture.

02 / Interface

Every module felt separate

A doctor could find the data, but moving between areas made the story harder to follow.

03 / Action

Some workflows stopped early

An alert or button could appear without showing what changed after the doctor acted.

The job was to make the next decision clear without hiding the evidence behind it.

Patient directory with profile images, clinical identifiers and view actions
Fig. 02The patient directory gives each person a recognisable identity while keeping clinical identifiers and actions easy to scan.
02

How I framed it

I stopped treating the home screen as another dashboard. Four questions decided what belonged in the first layer and what could wait.

Four linked questions organise clinical review: priority, patient context, change over time and source evidence. Choose a question in the diagram or list to inspect the related design decision. Previous and next controls, plus the arrow keys, move between steps.

Clinical decision flowFOUR QUESTIONS, ONE CLEAR NEXT DECISION01 · PRIORITYPriorityThe worklist brings theconcern, urgency and next02 · CONTEXTContextIdentity, allergies andcare history stay visible03 · TIMELINETimelineOne continuous historymakes change easier to04 · EVIDENCEEvidenceSource records remainavailable for verificationCURRENT LENS · PRIORITYWhat needs my attention now?The worklist brings the concern, urgency and next action together.

Step 1 of 4: Priority

I reviewed the existing screens, traced incomplete actions, compared clinical interface patterns and tested the rebuilt flows in the local React app. Where I did not have clinical usability data, I treated the outcome as a design direction rather than a proven impact.

Evidence rule: this case study only reports changes visible in the portfolio build and its test suite. It does not claim clinical accuracy, adoption, or time saved.
03

The product decisions

The same rule guided every screen: compress the signal, then let the doctor expand the context and the source evidence.

Decision 01

Make the home screen a worklist

The original cards were too wide and too slow to scan. I changed the page into a compact queue, ordered by clinical priority.

Repeated fields now stay in columns. Patient photos help recognition, while text and status labels carry the clinical meaning.

Notification inbox with aligned icons, patient context, priority and follow-up actions
Fig. 03The notification inbox uses the same triage logic: aligned signals, enough patient context, and one clear follow-up action.
Decision 02

Keep safety context in one stable place

Patient identity and allergy status now begin every patient workflow. The main action sits with the page heading, and the safety area no longer floats at the edge of the layout.

The same structure carries into medication review, so the doctor never has to relearn where critical context lives.

Patient overview with identity, allergy safety context, attention items and clinical summary
Fig. 04Overview starts with identity and safety, then moves into attention, context and the next decision.
Medication review using the same patient identity and allergy safety structure
Fig. 05Medication review keeps the same patient header and makes the review action visible at the top.
Medication review dialog with dose, schedule, indication and three required safety checks
Fig. 06The review cannot be completed until the doctor confirms the dose, the indication and the allergy check. The action now has a real finish state.
Decision 03

Make time feel continuous

The first timeline used separate date blocks. It looked tidy, but the dividers interrupted the chronology. I rebuilt it around one vertical spine.

The date is now part of each event. The doctor can read down the history without cross-referencing a detached column.

Continuous vertical clinical timeline with dates attached to each medical event
Fig. 07One uninterrupted spine preserves the sequence, while filters narrow the type of event.
Decision 04

Bring the signal closer to the list

Seeing the direction of one biomarker originally took three interactions. The rebuilt version brings the latest result and its direction into the list itself.

3
interactions · before

Open the item, switch tab, open the graph.

0
interactions · demo

Latest result and direction, already in the list.

Labs and trends interface showing latest biomarker results and direction inside the list
Fig. 08The primary list shows the latest result and its direction. A deeper comparison graph remains a proposed next layer.
Decision 05

Show review state before sharing

A generated medical report still needs clinical judgement. The list makes processing, review and completion visible instead of treating generation as the finish line.

This case study does not reproduce source report content. It shows the workflow and keeps sensitive-looking evidence out of the portfolio.

Clinical reports list with processing, review and completed states
Fig. 09Review state stays visible in the list, so report generation never looks like automatic clinical approval.
Decision 06

Make every action finish the job

Several actions used to end with a generic toast. In the demo, acknowledgement, review, notes and restored attention items all change the visible state.

Source records open from the table, and Settings changes the local interface without an account, a password, or an external service.

Medical records table with source document and review actions
Fig. 10Source evidence stays one clear action away from the medical-record table.
Local-only settings screen with profile, notification, density and text controls
Fig. 11Settings uses one compact shell, and every control changes the local experience.
Original medical record viewer with document preview, source metadata and a close control
Fig. 12The original-record viewer keeps source context in a focused document stage, so the doctor can verify extracted data without losing the patient workflow.
04

The report was the end product

The platform was designed to move from scattered source records to one clinician-reviewed health report. That report had to carry the evidence, surface the important findings, and still make it obvious that a doctor was responsible for the final interpretation.

I explored eight report directions against a 28-point review checklist. The direction below became the clearest end-to-end expression of the product: summary first, trends and detailed results second, then recommendations and follow-up.

Interactive report prototype · scroll insideOpen full report ↗
ArtifactA scrollable, fictional report showing how extracted records become a reviewable clinical narrative.
05

What changed

By the end of the project the product had moved from disconnected Bubble screens to a reusable React system built around one doctor workflow. Patient review, clinical records and report generation became one thing instead of three separate tools.

10
product areas designed

Seven clinical modules, plus patient management, billing and onboarding.

28
checks in the report framework

A repeatable quality bar for structure, graphs, evidence and clinical readability.

26
platform documents

Five sections that gave design and engineering a shared product reference.

Also reported by the project: reusable React layouts and closer design-to-code collaboration saved roughly 50% of development time. The original calculation was not in the reviewed archive, so this remains a team estimate.
06

What the team said

The work moved quickly, but the feedback I value most was about the quality of the thinking, and about how closely design and implementation stayed connected.

“Outstanding job! In just one week, it felt as though we had been working together for over a month.”
Jasper MiddendorpEngineering Lead, N1.care · Shared over Slack while working together
“Dhrumil is an exceptional UI/UX designer and Framer developer with meticulous attention to detail. He delivered expertly crafted, highly interactive designs that were extremely thoughtful from a usability standpoint. His Framer expertise is extensive and practical, allowing him to execute complex ideas quickly and precisely. His communication was clear and professional throughout the engagement and the quality of work consistently exceeded expectations. He is also highly skilled in AI platforms, including Claude Code and Gemini, and brings a modern and strategic edge to his work. Highly recommended to anyone seeking top-tier design talent.”
Jean-Yves SireauFounder and CEO, N1.care · Public on Upwork
07

What changed in my thinking

I stopped designing pages and started designing decisions. That changed how I chose components, wrote labels, positioned actions and judged density. A clean page still fails when the next decision is unclear. A working button still feels broken when nothing visible changes.

The most useful design moves were usually the plainest ones: keep safety context stable, attach dates to events, align repeated fields, and show what happened after an action.