AI product

AI agent

web app UI

human-AI Interaction

react

SPARC-P — AI-Powered Clinical Communication Training for Pediatric Clinicians

App Design

Project

Dashboard Design, AI Interface Design, Front-End Development, Web App Design

Timeline

12 months

What I did

End-to-end design and development of a web-based AI dialogue platform for clinical communication training, bridging character design, conversational AI, and learning experience into an accessible interface

Team

Character Designer, Backend Developer, AI Engineer, Learning Experience Designer

Overview

SPARCP is a AI training tool for doctors & nurses to practice evidence-based clinical communication.
‍
SPARC-P helps pediatricians build confidence in discussing HPV vaccination with hesitant parents through AI-powered conversational practice.Vaccine counseling is one of the most challenging aspects of pediatric care—clinicians must balance medical evidence with empathy, cultural sensitivity, and trust-building, often in high-pressure appointment settings.

background and Problem

Pediatric clinicians frequently face high-stakes vaccine conversations with hesitant or misinformed caregivers.

Why is this training needed?

Clinicians often struggle with sensitive, real-time conversations with patients.

Traditional training relies on limited role-play or one-time workshops.

There’s a need for scalable, repeatable, low-risk opportunities for clinicians to practice these conversations.

Research Question

"How might interaction design shape the quality of deliberate practice in AI-mediated clinical simulations?"

System Architecture

Meet the AI Agents

Three specialized AI agents operate in parallel: the Parent Agent (emotionally responsive roleplay), the Coach Agent (C-LEAR* aligned feedback), and the Supervisor Agent (safety, flow control, guardrails). My role was designing the interface layer where all three become a coherent human experience.

*C-LEAR Communication Framework

Counsel (recommend the vaccine)
Listen (invite concerns)
Empathize (acknowledge and validate)
Answer (address the concern directly)
Recommend (close with a strong recommendation)

IDEATION

Multiple Iterations

To meet tight deadlines, we prioritized rapid iterations, starting with high-fidelity mockups. We refined key areas for the AI Agent Training page:

Coach Feedback

Feedback display timing and location, keeping roleplay immersive

Context

Patient background and information. C-LEAR guidance

Interaction

Control buttons and feedback from the software

TESTING THE SOLUTION

User Testing and Stakeholder Review

To validate the designs, we conducted sessions with 6 participants to evaluate the platform's core interaction model using Iteration 1. We chose to test this iteration specifically because it introduced the video call–style interaction model: a deliberate design decision to mirror tools like Zoom that our clinician users already use daily for telehealth. The goal was to assess whether this familiar interaction pattern lowered the learning curve and made the simulation feel credible enough to practice in.We are using iteration 1 as we want to achieve the zoom-like interaction which most of our clinician users are comfortable with

Learning and Insights

Coach Feedback

Insight: Clinicians skipped or ignored the feedback blurb mid-conversation, treating it as a interruption rather than a resource.

‍Action: Explore call-to-actions to reframe feedback as forward-looking coaching rather than evaluation. Try adding a timed session before user can skip.

PHASE & PROGRESS INDICATION

Insight: Users felt disoriented across session phases, unsure which C-LEAR technique was currently expected and how far along they were in the session.

Action: Need to highlight the current phase and add progress indicator.

INTERACTIONS

Insight: Clinicians wanted to stay immersed in the patient conversation. Secondary controls and context panels created visual noise that broke simulation realism.
‍
Action: Move session controls to a minimal bottom bar modeled on video call conventions, making the interface feel like a real telehealth session. Patient background and C-LEAR reference can be moved either into a side panel or a dropdown.

Product Walkthrough

The Pilot

AI Training

Result & Next step

Current Phase

7 Participants

We're in the process of inviting clinians to test the product, we hope to scale the product and set up dashboard to view data analytics from sessions

Next Steps

The current simulation trains clinicians through structured roleplay. The next phase introduces a conversational AI coach that clinicians can talk to before entering a formal training session: ask questions, rehearse specific phrases, explore edge cases, or simply think out loud about a difficult parent scenario they encountered that week.
The proposed chat interface can make invisible process interactive and scaffolded. The design question this opens up is how AI should behave as a professional learning partner rather than a task executor. In the context of clinical training, that means the coach shouldn't just answer questions — it should model the reasoning behind good communication, surface relevant C-LEAR principles without being prescriptive, and know when to push back versus when to affirm. This positions SPARC-P as a research opportunity at the intersection of human-AI interaction and professional learning: how do we design AI interlocutors that support expert skill development, not just novice onboarding?

Rough prototype of AI Chat function