AI-assisted triage that speeds decisions without taking control away from nurses.
health-Tech · AI-assisted workflow · Safety-critical UX
Full app working
Problem
Classical hematology referrals range from benign abnormalities to early signs of malignancy. Triage nurses and hematologists must review fragmented referral notes, labs, and messages to decide urgency and disposition—often under tight capacity constraints.
Delays and inconsistencies in triage directly affect patient safety, access to care, and clinician workload.
Why it happens
Clinical data is fragmented
Labs, imaging, notes, and messages live across multiple EHR modules.
The cost of errors is uneven
Over-triage wastes limited clinic capacity. Under-triage risks patient harm.
AI trust is fragile
Black-box recommendations are unusable without clear reasoning and provenance.
The Goal
Design a triage assistant that standardizes the first pass of hematology referrals while preserving clinician judgment.
The system needed to reduce time to safe decision, improve consistency, and make AI reasoning transparent, auditable, and easy to override.
Users & their needs
Triage Nurse - Primary User
Pain: Manual review of referrals is slow and fragmented. Labs, referral notes, imaging, and messages are scattered across multiple EHR modules, increasing cognitive load and making prioritization difficult.
Consequence: High-risk referrals can be delayed, while low-acuity cases may consume unnecessary clinic time. Inconsistencies in triage decisions increase workload and safety risk.
Need: A calm, unified triage workspace that surfaces the most relevant data first, enforces required information, and provides explainable AI support without removing clinical control.
Doctors - Secondary Users
Pain: Referrals often arrive with incomplete or poorly organized information, requiring time-consuming clarification and re-review.
Consequence: Delays in identifying which patients require urgent face-to-face care versus those suitable for electronic consultation.
Need: Clear triage logic, visibility into how decisions were made, and confidence that high-risk cases are escalated appropriately while routine cases are handled efficiently.
Constraints & roadblocks
1
Safety and regulatory constraints: Clinical workflows operate under strict safety, privacy, and audit requirements.
Impact
AI could not act autonomously
Every action needed to be traceable and reviewable
Free-form data entry had to be limited
Design Decicions
Over-triage wastes limited clinic Read-only data sourcing from the EHR
Visible provenance, timestamps, and activity logs
AI outputs treated as drafts, never decisions.
2
Trust barriers around AI in clinical care: Clinicians are skeptical of AI that feels opaque or authoritative.
Impact
Black-box recommendations would be rejected outright
Overconfident language increased perceived risk
Design Decicions
Explicit explanation panels instead of hidden logic
Neutral, suggestive language in AI outputs
Clear visual separation between AI content and clinician decisions
3
Fragmented and inconsistent data: Referral data varies widely in completeness, structure, and quality.
Impact
Many referrals arrive missing critical labs or documentation
AI accuracy degrades with incomplete inputs
Design Decicions
Required data checklist.
High-impact actions gated until essentials were present.
Timeline and summary cards prioritized recency and relevance.
4
Cognitive load in high-volume triage: Triage nurses manage many referrals under time pressure, often on small screens.
Impact
Dense interfaces slowed decision-making
Important signals were easy to miss
Design Decicions
Calm visual hierarchy with limited color use.
Progressive disclosure for trends and draft notes.
Scan-first layouts, optimized for quick judgment.
Design solution
An integrated, nurse-first triage workspace that brings referrals, clinical data, and AI support into a single, explainable workflow.
Overall Design Decisions
Human-in-the-loop design
AI assists decision-making without replacing clinical responsibility.
Explainability built in
Every recommendation links to specific labs, referral text, and timestamps.
Safety is enforced, not suggested
Required data checklists gate high-impact triage actions.
Overrides are expected
Every recommendation links to specific labs, referral text, and timestamps.
Designed for real workflows
Built to mirror how triage actually happens in a medical center.
Feature 1 - For triage nurse
Triage dashboard
Workload overview with urgency cues, filters, and recent activity for fast re-entry.
fig 1.0 : Roadblocks addressed - 2, 3.
My key contribution here was pushing the conversation beyond the surface level. In
brainstorms, I helped the team design a simple workflow while accommodating key features needed for a nurse to make decisions.
Feature 2 - For triage nurse
Unified patient workspace
Labs, imaging, documents, and clinical timeline in one place with minimal cognitive load
fig 1.1 : Roadblocks addressed - 2, 3.
Feature 3 - For triage nurse
Privacy mode
Clear reasoning, source links, and an editable draft triage note.
fig 1.2 : Roadblocks addressed - 2, 3.
Feature 4 - For triage nurse
Gated Decision flow
Disposition options with required data checks and structured override pathways.
fig 1.3 : Roadblocks addressed - 2, 3.
Feature - For Doctors
Centralized the Referral Chaos
Nothing is hidden. Nothing is duplicated. No tab gymnastics.
fig 1.4 : Roadblocks addressed - 2, 3.
Before: Doctors hunt across systems to understand one referral.
Now: Each referral opens into a single workspace that brings together:
Referral note
Relevant labs and trends
Documents and attachments
Clinical timeline
Impact
The system demonstrates how AI can be safely embedded into clinical workflows by:
Reducing triage time without sacrificing judgment.
Increasing consistency across referrals
Improving trust through transparency and provenance
It moves AI from an experimental layer to a practical clinical assistant.
Whats next
The system demonstrates how AI can be safely embedded into clinical workflows by:
Reducing triage time without sacrificing judgment.
Increasing consistency across referrals
Improving trust through transparency and provenance
It moves AI from an experimental layer to a practical clinical assistant.
Explore more case studies
Comming soon
PromptR- AI-assisted prompt tool, to structure prompts for vibe coding websites
Agentic AI . VibeCoding
.
Responsive web
.
3 months