Designing The multi-omics AI platform behind QuantumLife's expansion into institutional healthcare.
Leading product strategy and design for a clinical AI platform adopted by two institutions in Asia.
Physician-in-the-Loop AI
Clinical Interface Design
Enterprise Admin Architecture
Deployed across Hong Kong & Japan
Cross-Regulatory Design (HK PDPO + Japan Ethics Committee)
0→1 Product Design
Impact
Launched at 2 of Asia's Top Private Healthcare Institutions.
IHH Healthcare
$5.3B revenue, 80+ hospitals across Asia
Hotel Chinzanso Tokyo
Japan's first multi-omics AI service
in regenerative medicine
Operates without on-site QuantumLife support staff. Critical for B2B SaaS unit economics.
From product discovery to first hospital deployment in 3 months.
01 Product Context
From Genomic Analysis Technology to Enterprise Clinical Platform
QuantumLife had developed AI-powered genomic analysis technology but no product, the technology remained uncommercial. Securing the company's first institutional customer required translating multi-omics AI into a platform that hospital procurement could approve, physicians would adopt within existing consultation workflows, and the business could scale without on-site support.
This translation posed three simultaneous constraints:
The platform had to satisfy institutional compliance requirements
Constraint 1
Integrate into existing clinical workflows rather than impose new ones
Constraint 2
Remain viable for a startup with limited engineering resources
Constraint 3
02 Leading Product Strategy and Interface Design
I led product strategy and end-to-end interface design as the designer on the project. For Longevity.Omics, I independently drove:
✍️Product Strategy
🔍UX research
🧩IA (Information Architecture) Design
🏥Clinical Interface Design
👤Enterprise Admin System Design
📇Design System
🛠️Developer Handoff and QA
I worked directly with QuantumLife's CEO and engineering team, translating business requirements and physician research into a deployable platform within three months.
03 The Challenge: Stakeholder Constraints
Multi-sided platform design: One platform must work for hospital administrators, physicians, and a resource-constrained startup, at once.
04 Architectural Pivot
From a Unified Portal to Mirroring Hospital Reality
The initial approach followed standard platform logic: a unified physician portal consolidating patient management, test ordering, report review, and billing into a single interface.
The prototype was tested at partner hospitals. In the second round of validation, one piece of feedback surfaced: a physician paused during the payment module walkthrough and said, "We don't handle payments."
This feedback revealed more than a missing feature. If physicians don't handle payments, then someone else at the hospital does. The portal was organized by software function, but hospitals organize around roles. It was built on an incorrect model of how hospitals actually operate.
So I think，understanding how hospitals distribute responsibilities across roles became the basis for the design process.
05 Institutional Research
Mapping Administrative-Clinical Separation Across Hospital Types
The role boundary identified in testing was not a local preference. It reflects how healthcare institutions organize themselves.
In large hospitals, entire departments handle finance, billing, procurement, and vendor management.
Even in clinics with one or two administrative staff, the person handling billing is not the physician. The scale compresses, but the boundary stays.
Critical Discovery
Despite different scales and role complexity, the separation between administrative operations and clinical work is universal.
The platform had to reflect this boundary at the architecture level. Two distinct workspaces sharing infrastructure, instead of a permission layer over a unified interface.
06 The Solution: Role-Based Architectural Separation
One Architecture, Two Portals, Every Hospital Size
The institutional research directly informed the architectural restructure. Rather than a unified interface with configurable permissions, the platform separated administrative functions from clinical workflows at the foundational level — mirroring the operational separation inherent in hospital organizations.
Clinical Workspace (Doctor Portal)
Operational Layer (Admin Portal)
This solved the three-way constraint:
For Physicians
A clinical workspace containing only clinical work. Patient data, AI analysis, and clinical decision tools.
For institutional operations
A configurable operational layer. Whoever handles non-clinical work at each institution uses it. Account management, billing, procurement, oversight.
For QuantumLife
One architecture scales across institution sizes. The operational layer expands or compresses. The clinical layer stays invariant.
Trade-off
Building two workspaces increased initial development scope. The operational layer (sub-account management, payment tracking, configuration) required engineering investment a unified architecture wouldn't.
Outcome
One architecture serves every hospital size.
Large hospitals get full institutional oversight through a populated operational layer. Boutique clinics run with a compressed version. The clinical workspace stays identical across deployments.
The architecture also supports feature-level modularity. Gleneagles manages ordering through their internal hospital system, so the Gleneagles deployment removes the Order function from the doctor portal entirely. Smaller clinics without internal ordering systems retain it. The platform adapts to each institution's existing operational infrastructure rather than imposing a fixed feature set.
07 Clinical Interface Design
Designing for the Physician-Verified Consultation Workflow
The clinical interface is the product's highest-stakes surface. Physicians use it during live consultations to verify AI output, contextualize findings, and communicate results to patients. Whether they trust and adopt it determines whether hospitals renew contracts.
Before this product, no established category covered this workflow. Sequencing labs delivered raw data. Direct-to-consumer platforms automated reports without physician judgment. Manual consultation preserved clinical authority at the cost of scalability. Longevity.Omics positioned AI analysis inside the physician's workflow, not as a replacement for it.
Every interface decision in this section operates under three simultaneous constraints:
Compatibility with existing clinical consultation workflows
Constraint 1
Preservation of physician authority over AI-generated content
Constraint 2
Completion within a standard 15-20 minute consultation window
Constraint 3
07.1 Gen-Decoder — Dual-Panel Validation Workspace
Sequencing labs produce reports containing tens of thousands of genetic variants in highly technical terminology. Most patients can't interpret them. Most physicians can't either, not without specialized genetics training.
A single person carries 4-5 million variants. Multiple variants often point to the same clinical outcome. Cross-referencing them across a full dataset exceeds reliable human capacity, even for specialists.
This is what QuantumLife's AI is built to do. It performs the cross-analysis that human cognition cannot do at scale, synthesizing variants across multiple omics datasets to surface clinically meaningful patterns.
AI analysis alone, however, is not a clinical product. Patients need a physician to contextualize what findings mean for their specific situation, decide what requires intervention, and explain it in terms they can act on. QuantumLife's model is physician-verified. AI handles the analysis that exceeds human cognitive capacity. Physicians provide clinical judgment and patient communication.
The interface for Gen-Decoder Report sits at this intersection. It must do three things simultaneously:
Accuracy
The actual patient report, not only a preview
Efficiency
Complete verification within a 15-20 min consultation
Trust
Traceable reasoning behind every recommendation
The interface uses a dual-panel workspace. Left panel: the patient's actual report — the document they will receive. Right panel: AI analysis organized for physician review. The physician reads the AI's interpretation on the right while seeing exactly what the patient sees on the left.
Accuracy
The left panel displays the actual patient-facing report. Physicians verify output quality during the consultation workflow. In a model where physicians stake professional credibility on AI-generated content, real-time verification is a prerequisite for clinical adoption — not a feature enhancement.
Efficiency
The right panel distills complex genetic findings into scannable clinical summaries. Each condition surfaces a brief clinical description, risk factors, and disease classification — eliminating raw variant codes from the physician workflow. Physicians assess the clinical picture without interpreting sequencing notation, completing review within consultation time constraints. The interface needed to outperform the alternative workflow, not just improve it.
Color coding distinguishes risk factors, prevention strategies, and attention items at a glance. Variant locations and their clinical consequences are labeled concisely for quick scanning.
Trust - AI credibility
The action plan interface is editable by default. AI-generated recommendations appear as a starting point, not a final output. Physicians modify dosage, timing, and priority based on patient specifics. The design signals that AI output is a draft in a clinical workflow, subject to physician review before it reaches the patient.
Trust - Clinical credibility
A patient with a genetic skin condition and a recent broken leg requires different immediate guidance than the AI's default recommendation. Physicians can contextualize plans to account for current clinical state, co-morbidities, or patient preferences. Editable action plans are the mechanism through which physician judgment is preserved within an AI-assisted workflow.
The Action tab provides an editable default recommendation plan — physicians modify based on patient context, with a dedicated Note area for clinical annotations.
07.2 Omni-Health — Multi-Dimensional Clinical Decision
Gen-Decoder and Epi-Insight provide diagnostic analysis. They identify genetic risks and biological aging status. Diagnosis, however, is not decision. Longevity medicine requires prescriptive guidance: which risks demand immediate intervention, what actions optimize healthspan, how interventions should be prioritized and monitored.
Omni-Health synthesizes multi-dimensional patient data into a comprehensive clinical action strategy. Lab results, Gen-Decoder findings, Epi-Insight aging assessments. These inputs require a coherent reading path. The challenge was both informational and spatial: how to structure complex data so the physician's eye and mind arrive at the right decision at the right moment.
Four architectural principles:
Progressive Complexity:
Separate diagnostic inputs from prescriptive outputs. Clinical and Genomic tabs position first, establishing patient baseline through uploaded laboratory results and static genetic risks before AI-driven analysis.
Intelligent Filtering:
Interface selectively surfaces clinically actionable findings, filtering comprehensive datasets to clinical priorities.
Contextual Integration:
Each tab provides a focused view while maintaining connection to the complete patient picture.
Flexible Navigation:
Tab sequence establishes logical first-consultation workflow. However, interface supports non-sequential access for follow-up consultations.
Tab sequence and specific design decisions：
Clinical data comes first because it's the baseline physicians already know how to read. Lab results, imaging findings, ongoing health context. No AI interpretation needed.
Genomic data comes second because it provides static risk context. Monogenic diseases, carrier status, polygenic risks. Still descriptive, not prescriptive.
Both tabs establish what the patient is before any tab attempts to say what to do about it.
Clinical & Genomic tabs: Progressive Complexity
Color hierarchy guides attention: Red (accelerated aging) > Green (favorable) > Gray (normal). System displays only three categories: Overall System, Favorable organ, Accelerated organ. Physicians scan priority systems instead of reviewing every organ.
Epigenetic tab: Intelligent Filtering
Two-tier prioritization. High-priority items (immediate dose adjustments) visually separated from moderate-priority items (consider during relevant prescribing). Prevents alert fatigue while ensuring nothing safety-critical is missed.
Pharma tab: Intelligent Filtering
Positioned last deliberately. The AI-generated 6-Pillar intervention plan synthesizes insights from all preceding tabs. Physicians see recommendations only after they have reviewed the full clinical picture. All fields are editable. Physicians modify recommendations based on patient context, compliance likelihood, and resource availability. The plan is a starting point for clinical judgment, not an output to approve.
Action tab: Contextual Integration
The tab sequence follows the physician's existing clinical reasoning order — current health status, then inherited risks, then aging markers, then medication implications, then comprehensive action plan. This mirrors how physicians already structure longevity consultations, supporting their existing cognitive workflow rather than imposing a new sequence.
Direct access to any tab . This supports both comprehensive initial assessments and focused follow-up monitoring, where physicians access only the relevant dimensions without repeating completed reviews.
Flexible Navigation
Existing clinical decision support tools provide siloed analysis. Genomic risk, aging assessments, drug interaction databases live in separate systems. Physicians consult each independently, then synthesize mentally.
Omni-Health integrates these dimensions into a single decision workflow. Diagnostic outputs become inputs to an intervention strategy, read in one session.
08: Supporting Infrastructure
Standard Enterprise Patterns for Operational Independence
Beyond the core clinical interface, the complete operational infrastructure required design: patient management, service ordering, and report management for the Doctor Portal; dashboard, sub-account management, and payment tracking for the Admin Portal.
These interfaces employ standard enterprise patterns — data tables, search filters, status indicators — intentionally. Design restraint reserves cognitive novelty for the clinical decision support interface, where data complexity genuinely demands new solutions. Standard patterns also minimize training overhead — critical for a platform designed to operate without dedicated QuantumLife support staff at each deployment site.
BUSINESS IMPACT
09: Deployment Outcomes
From Product to Production
Gleneagles Hospital Hong Kong IHH Healthcare — USD $5.3B revenue, 80+ hospitals across Asia
Longevity.Omics became QuantumLife's first institutional deployment. The workflow-boundary architecture allowed Gleneagles to integrate the platform without disrupting their existing clinical workflow. Launched Q1 2026.
N2 Clinic Hotel Chinzanso Tokyo — Kokkikai General Incorporated Association
The second deployment validated cross-regulatory portability. The architecture navigated Japan's ethics committee approval, complied with both Japanese privacy law and Hong Kong PDPO, and adapted to Japan's fee-for-service medical culture. Pricing: ¥1.32M (domestic) and ¥1.65M (international residents) per patient. Launched April 2026.
What the deployment validated:
Cross-Institutional Portability
The same workflow-boundary architecture runs at Gleneagles (large IHH hospital network) and N2 Clinic Tokyo (boutique regenerative clinic). Different scales, different countries, different regulatory regimes. No structural changes to the clinical workspace.
Architectural Flexibility
Gleneagles uses their own internal ordering system. During deployment we removed the Order function from the clinical workspace and integrated with their existing infrastructure. The operational layer continued handling physician account management. Clinical workspace stayed unchanged.
Operational Independence
Both deployments operate without on-site QuantumLife support staff. The admin layer enables institutional self-management, a prerequisite for scalable unit economics across markets.
Longevity.Omics went from a set of AI models without a product to two institutional deployments across two countries in roughly a year. What shipped is not a clinical report generator. It's an architecture that accommodates how different institutions actually work.
Design, in this project, meant deciding where to place the boundary. Between AI and physician. Between clinical and operational. Between what stays invariant and what adapts per institution. Every architectural decision was a boundary decision.