Back to Portfolio

Project overview

Designing decision clarity in clinical trial systems

UI/UX design · Adobe designathon

We redesigned a clinical intelligence workspace for faster, defensible decisions under time pressure. The sprint centered trust, transparent AI signals, and a calmer hierarchy than typical enterprise dashboards. The full flow was built end-to-end in 10 hours with Hongxin Li under rapid critique loops.

Timeline

10 hr build · April 2026

Role

UX / UI · Strategy

Tools

Figma · Adobe Express · Cursor

Team

Violet Hyun · Hongxin Li

Altrovia clinical trial intelligence dashboard mockup on a Studio Display

Mockup · High-fidelity hero from the designathon submission. Other sections on this page still use placeholders until export assets are added.

Problem discovery

Amendments cost teams before protocols ship

$43M

avg amendment cost per trial

$20M

estimated preventable exposure

75%

of trials require amendments

Poor benchmarking

Weak baselines inflate risk signals.

No precedent data

Teams guess instead of comparing cohorts.

Opaque AI

Outputs arrive without lineage or confidence.

Our approach

From problem list to design principles

Problems

  1. 1No defined target user
  2. 2Unclear goals per workflow stage
  3. 3Weak product differentiation
  4. 4Visual system lacked clinical authority

01

Confidence

Users need certainty from dense data, not more widgets.

02

Clarity over depth

Design for decision clarity and trust over feature count.

03

Right signal, right moment

Surface what matters now; guide the next action.

04

Clinical structure

A disciplined system that reads as serious, not consumer.

"Design for decision clarity and trust, not for more features or more information."

Strategy

How do we build the trust?

01

Competitor analysis

Mapped Medidata, Veeva, Palantir, and product references.

Captured strengths and gaps in a structured pros/cons chart.

02

Key insight

Trusted tools show their reasoning.

Palantir, Cursor, and Supabase never hide the machine.

03

Design principles

Surface AI visibly; never obscure how a value was generated.

Give review and override at every sensitive step.

04

Design direction

Sharper borders and teal for AI-augmented zones.

Layouts modeled on analyst tools, not marketing sites.

Outcome

A shared strategy: build trust through structure, transparency, and control—validated with stakeholder interviews and feedback from current users.

UX research

From scan to synthesis to wires

01

Competitor analysis

Mapped references

02

Adobe Express

Pros / cons chart

03

Wireframes

Clinical patterns

Annotated boards

Two annotated scans—clinical enterprise vs. product-marketing references—were reorganized in Adobe Express into the four-quadrant pros / cons chart.

Annotated competitor scan: Medidata, Veeva, and Palantir marketing and product pages with green and red notes
01 · Clinical platforms · Medidata · Veeva · Palantir
Annotated references: Linear, Supabase, Anduril, Scale.ai, Bunker, Vercel, and Palantir AI landing heroes with notes
02 · Product / dev landings · Linear · Supabase · Anduril · Scale · Bunker · Vercel · Palantir AI
Adobe Express

Typography, crops, and callout colors were aligned so both boards could collapse into one legible comparison.

Pros and cons synthesis grid: branding mismatch, bad visual systems vs. readability, credibility, and related notes
03 · Synthesis · Cons / pros matrix from both boards

Reference set

MedidataVeevaPalantirLinear.appSupabaseAndurilVercelBunkerPrimer.aiScale.ai

Takeaway: the most trusted products show their reasoning—so we tightened hierarchy, surfaced AI lineage, and borrowed navigation patterns from Linear and Palantir.

Visual system

Color and type

Palette

#578186

Altrovia teal

Primary actions

#EAEFF0

Teal light

Nav / hover

#F7F8FA

Surface

Cards

#151819

Dark

Primary text

#6B7280

Mid

Secondary text

#E5E7EA

Border

Dividers

Typography

Display — trial overview

Section header · active module

Button · AI badge · table label

Body copy for dense tables and helper text.

Caption · meta · timestamps

Process

How we built it

Process montage: exploration thumbnails and wireframes, dashboard UI, and Cursor editor with code and diffs
From exploration grids and wireframes to dashboard UI and Cursor implementation

01

Moodboard

Figma

02

Dashboard

Figma

03

MCP revisions

Figma MCP

04

Prototype

Cursor

05

AI assist

AI

Challenge

UI rules drifted across eight pages.

We cleaned up manually in Figma and Cursor—fast, but not perfectly consistent without a final audit pass.

Design challenges

Where the sprint got hard

01

Understanding real user needs

No direct access to end users.

  • Deep pass on the existing product and company context.
  • Structured questions to Chris on goals and pain points.
  • Used stakeholder input as a proxy for interviews.

02

Maintaining UI consistency

AI speed without precision.

  • Ran batches through Cursor to auto-fix drift.
  • Still needed manual polish on edge spacing.
  • Good enough workflow, not a perfect pipeline.

User feedback

What we heard

Emily Shen

NYU alum · Product designer, Palantir

"Rounded corners feel too sales-y. A serious clinical product needs sharper borders for authority."
Video call screenshot from feedback session with Emily Shen

Key feedback

  • Sharpen corners and grid rhythm.
  • Borrow analyst-grade density.
  • Reduce decorative chrome.

Christos

Altrovia AI

"The dashboard should read as a portfolio: clients see status and risk at a glance."
Video call screenshot from feedback session with Christos

Key feedback

  • Stronger emphasis on active trials.
  • More visual hierarchy than tables alone.
  • Client-ready overview first.

Final UI

Seven core screens

Altrovia workspace: trial table, search, filters, and pagination

01 · Workspace

Overview · trial cards · new trial CTA

Design parameters for TALAPRO-2 with AI-extracted badge and expandable protocol sections

02 · Design parameters

AI badge · collapsible rows · label/value rhythm

Trial analysis hub listing saved intelligence runs with scope tags and view actions

03 · Run analysis

Saved runs · scope badges · timestamps

Trial intelligence analysis: KPIs, feasibility cards, enrollment curve, benchmark table

04 · View results

Comparator table · amendment callouts

Generate protocol flow with readiness metrics and ten-section outline

05 · Generate protocol

Status bar · ten sections · readiness CTA

Version gallery with latest draft card and chronological protocol history

06 · Version gallery

Latest draft · history notes

Review extracted data step: study basics form and live complexity score footer

07 · Complexity assessment

Wizard · review · score bar

Demo

Prototype walkthrough

Reflection

What we fixed

We closed the biggest trust and layout gaps first: clearer AI provenance, consistent sidebars, and scanning patterns that felt clinical instead of consumer.

AI trust was invisible

Added extraction badges and source callouts on parameters.

Sidebar overlap

Rebuilt footer alignment across sidebars in one coordinated pass.

Consumer tone

Shifted to sharper borders after Palantir-informed critique.

Hard-to-scan parameters

Collapsible rows with muted label vs. dark value.

Reflection

If we had more time...

Design consistency

Full grid audit for spacing and alignment across every screen.

Landing page

Stronger clinical value story with real marketing visuals.

User testing

Sessions with trial ops to validate AI trust patterns.

Dashboard visuals

More emphasis on active trials and status, beyond tables.

It was also our first time taking a designathon from building wireframes to mockups, and we worked really hard and I am proud that we still brought something we felt good showing. When everything moves this fast, you try to use AI efficiently, not just to skip the hard thinking. I know more tight deadlines will come in real-life work settings, and this felt like good practice for that.

Continue

View all projects