Designing a human-centered clinical decision support experience grounded in physician research and AI safety requirements
Project Snapshot
Product: Lumia IA — clinical decision support assistant (concept-to-design)
Timeline: Dec 2025 – Jan 2026
Role: Lead UX/UI Designer (end-to-end)
Stakeholders: AI specialists (requesters) + practicing physicians (interviews/feedback)
Scope (What I designed): trust-first UI (sources/confidence/limitations), review-and-confirm workflow, auditability, safe language, reusable patterns for scalability
Outcomes (validation, illustrative): ~25–30% faster time-to-orientation; +15–20% perceived trust uplift
Confidentiality: Concept + early validation; metrics illustrative.
Summary
Lumia IA is a concept for a clinical AI assistant designed to support physicians during diagnostic decision-making.
In clinical environments, time is limited and decisions carry real consequences. Physicians don’t need more information — they need clarity, context, and confidence.
The challenge was to design an AI-assisted experience that speeds up understanding without removing control, keeping the decision clearly human while making reasoning transparent.
I led the UX/UI design end-to-end, working with physicians and AI specialists to shape an experience where AI supports thinking, rather than replacing it.
Context
Physicians often work under time pressure, navigating incomplete information and multiple possible diagnoses.
In this context, an AI assistant can either reduce cognitive load — or increase it.
If suggestions are unclear, overly confident, or disconnected from evidence, trust drops immediately.
Designing for this environment means respecting how doctors think: quickly scanning, focusing on signals, validating assumptions, and documenting decisions.
My Role
Lead UX/UI Designer working with the AI specialists who requested the project and interviewing practicing physicians to ground the product in real workflow needs.
The Design Challenge
The core challenge was not presenting AI suggestions — but making them usable in a clinical context.
The experience needed to:
- support fast understanding without oversimplifying
- make reasoning visible without overwhelming
- communicate uncertainty without reducing trust
- keep the physician fully in control
Constraints and approach
Instead of designing around “AI output,” I designed around the clinician’s mental flow:
skim → focus → verify → decide
The interface was structured to follow this sequence, ensuring that information appears at the right moment, in the right level of detail.
At the same time, AI constraints such as explainability, traceability, and safe communication were embedded directly into the interaction model.
Discovery and research
I grounded the design in two perspectives:
Physician interviews
- Interviewed physicians to understand how they confirm diagnoses, where they lose time, and what would make an assistant genuinely usable.
- Key themes (illustrative but realistic):
- Fast access to the “why” matters more than the final answer
- Clear separation between patient facts vs AI interpretation builds trust
- Doctors want assistance with summarization, differential hypotheses, and red flags—while keeping authorship of the decision
AI specialist interviews (project requesters)
- Interviewed AI specialists to define constraints for safety, explainability, and responsible use.
- Key themes:
- Traceability is required for internal review and governance
- Confidence must be communicated carefully (avoid false certainty)
- The interface needs structured feedback loops to improve model behavior safely
Key improvements (iteration highlights)
1. Clinician-first information hierarchy
The interface prioritizes what matters in sequence:
- Patient overview with key signals
- Clinical flags and missing critical data
- Diagnostic hypotheses (not conclusions)
- Evidence and reasoning linked to each suggestion
The goal was to reduce the time to “understand the case” without hiding complexity.
2. Suggestions designed as hypotheses
Instead of presenting answers, the system presents possibilities.
- Confidence shown as ranges, not certainty
- Clear “why” behind each suggestion
- Evidence always accessible at decision moment
This avoids overtrust while supporting faster reasoning.
3. Review-and-confirm interaction model
The workflow mirrors real clinical thinking:
- Compare hypotheses
- Validate evidence
- Confirm a direction
- Document reasoning
The system supports decisions — it doesn’t make them.
4. Feedback loop without friction
Feedback is embedded into the flow:
- Quick “helpful / not helpful”
- Option to flag issues
- Lightweight structured input
Improving the model without slowing the clinician down.
5. Visual language: calm, clinical, human
The UI avoids “AI aesthetics” and focuses on trust:
- Clean hierarchy and restrained colors
- Soft visual elements to reduce tension
- No “magic” or exaggerated AI signals
The system feels reliable, not experimental.
Approved layout
Final layouts designed by me, reviewed with AI specialists and validated through physician feedback sessions.
Outcomes
Validation sessions showed faster understanding of clinical cases and increased confidence when reasoning and limitations were visible.
Physicians reported feeling more in control of decisions, even with AI assistance.
What this demonstrates
This project demonstrates my ability to design AI-assisted products in high-stakes environments.
It shows how I balance usability, trust, and responsibility — translating complex systems into experiences that support human decision-making without removing control.