ClarityCare
A bare risk score is a black box — the real design work is engineering trust.
Care coordinators have to catch the patients most likely to fall through the cracks before the next crisis — but the tooling gives them an undifferentiated wall of EHR data and no idea where to start. ClarityCare is a concept for turning an AI risk model into a workflow a clinician can actually trust and stay in charge of.
“I know I’m missing someone — I just can’t see who.”
That’s the illustrative voice of the user: a coordinator staring at a wall of records with no time to guess. A risk model can rank the wall — but a bare score is a black box, and no clinician is going to act on a number they can’t interrogate. The core design problem is trust.
A concept project — the ML model, SDOH indices, and personas are assumed to make the argument concrete, not staffed or validated.
Risk isn’t only in the chart.
The model draws on more than the chart. The person’s circumstances outside the clinic are often the real driver, so the interface had to make those factors legible, not bury them. The inputs it folds together:
Turn the black box into a glass box.
Each feature answers a skeptical clinician’s question:
“Where do I start?”
The Smart Triage Queue ranks the wall of records so the riskiest patients surface first.
“Why should I believe this?”
Insight Cards render the model’s reasoning as plain language — “high risk due to: 2 missed follow-ups, 1 ER visit, unfilled meds.”
“Am I still in control?”
A Suggested Action + Override Flow keeps the clinician in charge — and a structured override doubles as the model’s next training signal.
Each model capability creates an interface obligation. The model’s honesty about its own limits is the UX.
The Smart Triage Queue answering “where do I start?” — the ranked wall the first question above describes.
Thirty seconds of attention in someone’s living room.
Mobile wasn’t a shrunk desktop — it was re-architected for the home visit: a coordinator with thirty seconds of attention, standing in a living room, needing to re-score risk on the spot after entering a new SDOH fact. Same system, a completely different center of gravity.
With AI in the loop, the interface’s job is to make the model accountable to the human using it.
The deliverable is the argument and the high-fidelity system that embodies it: a triage queue, an explainable insight card, an override flow that keeps the clinician in charge, real-time SDOH entry, and a timeline — a black box turned into a glass box.
A concept project with no clinician validation or live model — so there are no metrics, and I don’t invent any.