Signals without the full picture
A single measurement is difficult to interpret in isolation. Trends, personal baselines, recent activity, sleep, medications, and upcoming appointments can change the meaning of a signal.
Independent product design concept
HealthOS: Designing an agentic health experience
Exploring how an AI health agent could connect signals, personal context, and explainable insights into a more proactive, human-centered healthcare experience.
Tools & technology
Product Strategy
UX / UI Design
Agentic UX
Next.js 16
React
TypeScript
Tailwind CSS v4
shadcn / Base UI
Lucide React
Recharts
A dashboard is a window into the agent’s work.
HealthOS explores a shift from passive health dashboards to a personalized health operating layer: an experience where an agent interprets changing signals in context, explains what it notices, and helps a person decide what to do next.
Rather than treating AI as a chatbot bolted onto a dashboard, the concept connects the agent to a shared patient state. The dashboard, insight cards, evidence, and conversation all reflect the same selected patient and the same underlying context.
The experience is organized around a simple principle: Observe → Understand → Explain → Act. The agent can surface a potential area of attention, make its reasoning visible, and present a possible next step while leaving review and approval with the person.
Design thesis: Trustworthy agentic healthcare is not just about producing an answer. It is about making the answer understandable, showing the evidence behind it, and preserving human agency over what happens next.
Move beyond fragmented signals and isolated answers.
Personal health information can come from many places: wearable devices, measurements, medications, appointments, and clinical results. A collection of metrics may show what changed, but it does not necessarily help someone understand how those changes relate to their own baseline or what deserves attention.
A conversational assistant can answer questions, but a conversation alone may not provide a clear view of the underlying evidence, the patient context being considered, or the status of an action. A dashboard can display information, but it may leave the user to interpret the pieces themselves.
HealthOS investigates how these experiences could work together: a persistent view of the person’s health context, an agent that explains what it notices, and a transparent path from insight to action.
Answers without enough explanation
An agent needs to communicate what it noticed, what information informed its interpretation, where uncertainty remains, and which actions still require human review.
Insights without a clear next step
Identifying a potential concern is only part of the experience. People also need a comprehensible way to review options, approve an action, or continue monitoring.
Experiences that ignore individual context
The same interface must accommodate different health situations without forcing every person into the same priorities, data, or agent responses.
Design the agent and the experience as one system.
The central design decision was to treat patient context as shared state across the entire experience, rather than designing separate dashboards or independent conversations for each scenario.
A selectable patient context changes the information, metrics, agent insight, and relevant workflow together. This creates a consistent product model while allowing the experience to respond to different needs.
Experience model: Observe → Understand → Explain → Act. Each stage gives the user a clearer view of what the agent is doing and how to participate in the decision.
STEP 01
Observe
Bring together signals, trends, and personal baselines.
STEP 02
Understand
Interpret changes in the context of the selected patient.
STEP 03
Explain
Present the insight, supporting evidence, and uncertainty.
STEP 04
Act
Let the person review, approve, or continue monitoring.
Make the agent’s work visible
The dashboard acts as a window into the agent’s work. A dedicated agent insight surfaces the current area of attention, a patient-specific explanation, and a confidence indicator. The user can open the insight to explore the explanation and move through a simulated review workflow.
Connect conversation to the dashboard
The conversational agent is designed as part of the product experience rather than a separate destination. Suggested prompts help users explore changes and results, while the conversation remains grounded in the currently selected patient’s illustrative data.
This relationship is important to the concept: the user can ask a question in natural language, but can also return to the dashboard to see the larger context around the response.
Design for different patient contexts
Three fictional patient scenarios make the shared-state concept tangible. Switching patients changes the dashboard’s content and the agent’s focus, illustrating how one product framework can adapt without becoming three disconnected experiences.
Scenario 01
Jordan
Resting heart rate rises from 57 to 62 over approximately 10 days alongside reduced sleep and activity. The scenario focuses on contextual interpretation without presenting the change as an urgent signal.
Scenario 02
Maya
At pregnancy week 24, Maya’s context includes a prenatal appointment on Thursday, blood pressure in range, and prenatal vitamin adherence of 6 out of 7 days.
Scenario 03
Alex
Alex’s type 2 diabetes scenario includes overnight glucose above baseline, time in range of 71%, and an A1c value of 7.1% to demonstrate a different pattern of agent attention.
Turn the product model into a working interactive prototype.
HealthOS was implemented as a front-end prototype to demonstrate the relationship between patient context, dashboard content, and an agent-guided workflow.
The prototype uses a shared patient data model for Jordan, Maya, and Alex. Patient selection updates the displayed information and the agent experience, making the underlying product behavior tangible rather than relying on static mockups alone.
One product, multiple contexts
Patient information follows a consistent schema, allowing the interface to render the appropriate metrics, signals, baselines, appointments, and agent insight for the selected scenario.
A connected interaction model
The prototype demonstrates investigation, explanation, evidence review, approval, and monitoring states through local interaction rather than a live AI service.
A dashboard with a conversational layer
The interface combines structured health information with a conversational agent, keeping the broader patient context visible while the user explores an insight.
Real interface behavior
Built with Next.js, React, TypeScript, Tailwind CSS, shadcn/Base UI, Lucide React, and Recharts, using local state and illustrative data to power the experience.
Scope and implementation boundaries
The experience is intentionally a concept prototype, not a production healthcare application. It does not connect to an electronic health record, wearable device, real AI model, backend service, or scheduling system. Patient switching and agent workflow behavior are simulated with local data and state.
This scope made it possible to explore the product architecture and interaction model without suggesting that clinical intelligence, medical recommendations, or real-world actions had been validated.
A tangible foundation for exploring agentic healthcare UX.
The result is an interactive portfolio concept that demonstrates how a patient-aware agent could be integrated into a health dashboard through shared context, visible reasoning, and human-in-the-loop interaction.
Shared patient context
Three fictional scenarios use a common product framework, demonstrating how patient switching can change dashboard content and agent behavior together.
Explainable interaction
Agent insights connect to an explanation and review workflow, making the path from a surfaced signal to a possible action visible.
Conversational integration
The simulated chat experience is part of the broader dashboard rather than an isolated interface with no persistent context.
Functional front-end prototype
The concept is implemented in a working web interface, allowing the product structure and core interaction patterns to be explored directly.
What I would validate next
The next phase would focus on evaluating whether people can understand an agent’s explanation, identify the evidence behind an insight, and maintain an appropriate sense of control over next steps.
I would test the experience with patients and clinicians, explore how confidence and uncertainty should be communicated, and define clinical, privacy, and safety guardrails before connecting the concept to real health data or actions.
Key takeaway: Designing an agentic product means designing more than a conversation. It means defining the shared context, visible reasoning, interaction states, and human decisions that make the agent understandable and useful.
Project disclaimer: HealthOS is an independent portfolio concept inspired by healthcare experiences. It is not an official Baylor Scott & White Health product and is not affiliated with, endorsed by, or developed in partnership with Baylor Scott & White Health. All patient scenarios and data are fictional and illustrative. This prototype is not a medical device, does not provide validated clinical recommendations, and has not been clinically validated or formally user-tested.