Quantum
Critical clinical data, finally legible
// THE CONTEXT
Quantum turns fragmented patient records into a single intelligent longitudinal timeline, readable at a glance even twelve hours into a shift.
Year
2023
Client
Regional hospital network (confidential)
Role
End-to-end product design and engineering
Duration
9 months
Team
5 engineers, 1 product designer, 1 clinical consultant
Category
MedTech · Clinical Data Visualisation
In ICUs and internal medicine, clinicians juggle Epic, Cerner, and local spreadsheets just to reconstruct a patient's history. That fragmentation costs time, introduces errors of omission, and hides slow-moving dangers: a creeping creatinine rise, an unflagged pharmacogenomic interaction. FHIR standards already existed; no one had used them to deliver a unified, non-invasive view.
The challenge wasn't reading FHIR data. It was doing so reliably across two EHRs with divergent conformance profiles, without ever writing back to the source systems. A high-contrast UI that met clinical accessibility guidelines while preserving information density took dozens of ward-side iterations. And the AI agent had to surface real anomalies without triggering alert fatigue.
Zeklar built a Python FHIR normalisation layer that reconciles Epic R4 and Cerner STU3 into a single canonical model, with provenance tracked at the resource level. The timeline renders entirely in D3.js with semantic zoom: the deeper you go, the more detail emerges. The AI agent runs continuously over a rolling 90-day window, combining codified clinical rules (CDS Hooks) with a model trained for low-frequency anomaly detection. The UI is built around a dark palette calibrated to 6500K to reduce eye strain under hospital lighting.
// Results
Clinical impact, measured
−64%
Patient lookup time
91%
Alert precision
78%
30-day adoption
7
Hospital sites
−41%
Eye strain
340+
Interactions flagged
// What made it special
Non-invasive FHIR integration
Read-only connection to Epic and Cerner with no changes to existing workflows or contracts, deployed per site in under two weeks.
D3 longitudinal timeline
A semantically zoomable view plots medications, labs, diagnoses, and notes on a single time axis, with per-specialty filtering.
AI anomaly agent
A background agent analyses 90 days of history, surfacing only high-specificity anomalies to keep noise low and signal high.
Designed for 12-hour shifts
A 6500K dark palette, high-contrast typography, and density-optimised layout, validated against visual-fatigue measurements with clinicians.
// The process
Clinical discovery
Six weeks of ward shadowing to map real workflows, touched systems, and the highest-cost friction points.
Data architecture
FHIR normalisation layer, canonical internal model, and ingestion pipeline with an immutable audit trail.
Prototyping & testing
Biweekly iterations with clinicians on high-fidelity prototypes, including A/B tests on density and low-light readability.
Rollout & monitoring
Phased ward-by-ward rollout with 45-minute onboarding, then 8 weeks of active monitoring and AI agent tuning.