AI-Assisted Diagnostic Reasoning Interface
AI-Assisted Diagnostic Reasoning Interface
Designing Human-in-the-Loop AI for Complex, Uncertain Clinical Decisions
Designing Human-in-the-Loop AI for Complex, Uncertain Clinical Decisions
Role
Product Designer (AI + UX Systems)
Focus
Clinical decision support, temporal reasoning, AI transparency
Outcome
Designed a risk-aware AI interface to help clinicians reason through complex recurrent infections without over-relying on automation.
Problem
Recurrent infections are deceptively difficult to manage.
In veterinary and human medicine alike, patients may:
• Improve temporarily with treatment
• Relapse repeatedly
• Show inconsistent lab signals
• Present incomplete histories
Clinicians must reconstruct patterns across time, treatments, and environmental factors - often under time pressure.
Most medical systems store data as isolated entries, not longitudinal narratives.
The real challenge is not more data, but seeing patterns across time without jumping to false conclusions.
Design Question
How might AI support clinical reasoning in recurrent infection cases without creating false certainty or replacing professional judgment?
Design Philosophy
This system is built on restraint.
Principles
• AI suggests, clinicians decide
• Time is the primary interface
• Uncertainty must remain visible
• Silence is safer than misleading certainty
• Transparency builds trust
The system is intentionally non-diagnostic.
Context: Real-World Case
The design was informed by a real pattern:
Patient: Cony, 11-year-old Shetland Sheepdog
Condition: Recurrent E. coli cystitis
Duration: ~6 months
Pattern: Temporary improvement with antibiotics, relapse after cessation
Clinical ambiguity included:
• Possible endocrine involvement (Cushing’s suspected, later doubted)
• Low historical urination frequency
• Environmental changes (water source)
• No structural abnormalities detected
This created a classic high-uncertainty case.
Solution Overview
I designed an AI-assisted clinical reasoning interface that:
• Aligns symptoms, labs, and treatments on a timeline
• Surfaces recurrence patterns
• Ranks hypotheses by evidentiary support
• Communicates uncertainty explicitly
• Avoids diagnostic claims
The system acts as a cognitive partner, not a decision maker.
Impact
• Prevented false reassurance by explicitly surfacing missing or incomplete data in the user interface
• Increased user trust in system outputs by making uncertainty and data gaps visible and understandable
• Enabled more informed decision-making through clearer representation of complex medical information
Role
Product Designer (AI + UX Systems)
Focus
Clinical decision support, temporal reasoning, AI transparency
Outcome
Designed a risk-aware AI interface to help clinicians reason through complex recurrent infections without over-relying on automation.
Problem
Recurrent infections are deceptively difficult to manage.
In veterinary and human medicine alike, patients may:
• Improve temporarily with treatment
• Relapse repeatedly
• Show inconsistent lab signals
• Present incomplete histories
Clinicians must reconstruct patterns across time, treatments, and environmental factors - often under time pressure.
Most medical systems store data as isolated entries, not longitudinal narratives.
The real challenge is not more data, but seeing patterns across time without jumping to false conclusions.
Design Question
How might AI support clinical reasoning in recurrent infection cases without creating false certainty or replacing professional judgment?
Design Philosophy
This system is built on restraint.
Principles
• AI suggests, clinicians decide
• Time is the primary interface
• Uncertainty must remain visible
• Silence is safer than misleading certainty
• Transparency builds trust
The system is intentionally non-diagnostic.
Context: Real-World Case
The design was informed by a real pattern:
Patient: Cony, 11-year-old Shetland Sheepdog
Condition: Recurrent E. coli cystitis
Duration: ~6 months
Pattern: Temporary improvement with antibiotics, relapse after cessation
Clinical ambiguity included:
• Possible endocrine involvement (Cushing’s suspected, later doubted)
• Low historical urination frequency
• Environmental changes (water source)
• No structural abnormalities detected
This created a classic high-uncertainty case.
Solution Overview
I designed an AI-assisted clinical reasoning interface that:
• Aligns symptoms, labs, and treatments on a timeline
• Surfaces recurrence patterns
• Ranks hypotheses by evidentiary support
• Communicates uncertainty explicitly
• Avoids diagnostic claims
The system acts as a cognitive partner, not a decision maker.
Impact
• Prevented false reassurance by explicitly surfacing missing or incomplete data in the user interface
• Increased user trust in system outputs by making uncertainty and data gaps visible and understandable
• Enabled more informed decision-making through clearer representation of complex medical information
