📋 TL;DR Summary
The Problem
Healthcare providers globally, especially in resource-constrained environments, face severe shortages of specialists. Desktop-centric diagnostic tools fail to reach clinicians using smartphones as their primary device, leading to delayed diagnoses.
The Solution
Designed a multimodal medical AI platform that works offline, supports 25+ medical specialties, and transparently handles AI uncertainty through clear visual states and human-in-the-loop workflows.
My Role
Lead UI/UX Designer, spearheading the design of the platform from initial concept to a scalable design system deployed across international markets.
Business Impact
Achieved 97% diagnostic accuracy. Scaled to over 500 active facilities in 50+ countries (including the USA), generating $2.9M in annual revenue while reducing report generation time by 93%.
Moremi AI is a globally scalable multimodal medical AI platform deployed across 50+ countries, including the US. As Lead UI/UX Designer, I designed the end-to-end experience for an enterprise-grade HealthTech SaaS product that generated $2.9M in annual revenue (2024), served 500+ healthcare facilities, and was featured on CNN for its real-world clinical impact.
Business Constraint
Ghana had fewer than 80 radiologists for 33 million people. The platform needed to function in world-class hospitals and rural clinics using smartphones over 3G — a constraint that made this an enterprise infrastructure problem, not just a design challenge.
"How do you design for AI that's sometimes wrong? In healthcare, uncertainty isn't a bug — it's critical information."
The business requirements were non-negotiable:
- Match or exceed radiologist-level diagnostic accuracy (75–86% baseline)
- Work offline in environments with intermittent 3G connectivity
- Support clinicians ranging from tech-savvy radiologists to rural nurses with basic digital literacy
- Handle AI uncertainty transparently — clinicians needed to know when to trust the AI and when to escalate
Strategic Approach
1. Clinical-First Information Architecture
Restructured the platform around clinical workflows rather than technical AI capabilities:
Before: AI-centric navigation (Models → Algorithms → Results) After: Clinical workflow navigation (Patients → Diagnose → Review → Report)
This shift came directly from 23 in-depth clinician interviews and 15 contextual inquiries across Accra, Lagos, and Nairobi — clinicians think in terms of patients and diagnoses, not AI models.
2. Designing for AI Uncertainty
The most critical design challenge: handling cases where the AI isn't certain.
- Confidence Score Prioritization: Positioned the diagnosis confidence score at the top of the UI. User research revealed clinicians' first question was always "How confident is the AI?" — so I made it the first thing they see.
- Low Confidence Mode: When AI returns below 80% confidence, the interface shifts to a review mode with amber visual indicators, prompting specialist consultation before proceeding.
- Traffic-Light Uncertainty States: Green (high confidence) → Amber (review recommended) → Red (specialist required) — interpretable at a glance during time-critical situations.
- Human Override Patterns: Every AI recommendation includes a "Clinician Disagrees" action, ensuring the AI augments rather than replaces clinical judgment.
Designing for Trust
Use the slider to see how the UI communicates uncertainty.
Chest X-Ray Analysis
SCAN-ID: 8492-MA-2024
Analysis complete. High reliability.
3. Offline-First Performance Engineering
Collaborated with engineering to implement a Progressive Web App with local storage — core diagnostic functions work without internet connectivity.
- Progressive Loading States: Designed skeleton loading mirroring the API's three-stage response (image processing → model inference → confidence scoring), reducing perceived wait time by 40%
- Sub-2 second load times on 3G connections through aggressive image optimization and lazy loading
- Offline sync for rural healthcare settings with intermittent connectivity
4. Multimodal AI Interface System
Designed interfaces supporting 25+ medical imaging modalities (X-rays, CT scans, MRIs, mammograms), natural language clinical queries, and biological data processing — all unified under a single diagnostic dashboard.
Performance Benchmarks
Comparative analysis of diagnostic accuracy (AUC-ROC) between Moremi AI and board-certified radiologists. Data derived from the validation study published broadly.
Unified Diagnostic Dashboard:
- Single-screen view of patient information, imaging, and AI analysis
- Real-time confidence scoring for AI recommendations
- Side-by-side original vs. AI-annotated image comparison
AI Explanation Interface:
- Visual heatmaps showing AI focus areas on medical images
- Plain-language explanations of diagnostic reasoning
5. Automated Report Generation
Reduced report generation from 45 minutes to 3 minutes (93% improvement):
- Automated report generation with human review checkpoints
- Customizable templates for different medical specialties
- Multi-language support (English, Swahili, French, Yoruba)
- PDF export optimized for local printing capabilities
Automated reporting interface generating comprehensive diagnostic summaries with confidence scores and clinician editing capabilities.
Conversational AI interface processing medical queries with context-aware natural language processing.
Validation & Testing
Phase 1 — Concept Validation: 12 healthcare providers tested early prototypes. Task completion rate improved from 68% → 94% after design iterations.
Phase 2 — Clinical Validation: 3-month pilot across 5 hospitals in Ghana and Nigeria. 500+ diagnostic cases processed. 96% user satisfaction. 89% of providers reported increased diagnostic confidence.
Phase 3 — Scale Testing: Deployed to 50+ facilities with performance monitoring across device types and network optimization for low-bandwidth environments.
Business Outcomes
| Metric | Result | Context |
|---|---|---|
| Diagnostic Accuracy | 97% AUC-ROC | Pleural effusion detection (n=2,400) — outperformed 75-86% radiologist baseline |
| Cardiomegaly Detection | 90% accuracy | vs. 77–87% for individual radiologists |
| Report Generation | 93% faster | 45 min → 3 min per report |
| Facilities Onboarded | 500+ active | Across Ghana, Nigeria, Kenya (2024) |
| Operational Savings | 60% cost reduction | Average per facility, first 6 months |
| Platform Revenue | $2.9M annual | 2024 |
| UI Bug Reduction | 45% fewer | After design system implementation |
Recognition
- Forbes 30 Under 30: CEO Darlington Akogo featured for healthcare innovation (Class of 2025)
- CNN Coverage: Featured as breakthrough African AI platform
- 5+ peer-reviewed publications on AI healthcare applications
- M&A offer received (April 2025)
Design System Impact
Created a modular component library in Figma (WCAG 2.2 compliant), translated into a production React/TypeScript library. The shared design system reduced design-dev friction and enabled the platform's scaling from 3 West African markets to 50+ countries globally.
Impact Update (2025)
Since completing the UX design work, Moremi has scaled significantly:
- 50+ countries now using the platform (USA, India, Pakistan, Barbados, and more)
- 30 medical imaging modalities supported (up from 14 at launch)
- CEO appointed as African Union AI Strategy Consultant
The design system and UX foundation established during my tenure continues to support this global scale.
Key Learnings
- Trust Through Transparency: Showing AI reasoning process increased adoption by 300%
- Offline-First Design: Connectivity challenges require offline capabilities designed from Day 1, not bolted on
- Design for AI Uncertainty: AI interfaces aren't about automation — they're about augmentation. Designing "Low Confidence" states was the most impactful design decision in the entire project.
Impact Statement: Moremi AI demonstrates how thoughtful UX design bridges the gap between cutting-edge AI and real-world healthcare needs — achieving enterprise-grade results while remaining accessible to clinicians across the globe.