DYMOND
A Human-in-the-Loop Digital Phenotyping for Mental Health
Team
UbiWell Labs, Northeastern University, Dartmouth College
Skills
Mixed Methods UX Research, Wearable Device Usability Testing, Human-in-the-Loop AI Research, Co-Design & Participant-Centered Studies, Data Visualization & Dashboard Prototyping
Role
Researcher, Designer
Timeline
44 Weeks, Summer 2024 – Spring 2025
Designing Wearable Experiences for Mental Health Monitoring
DYMOND (Dynamic Monitoring for Depression) is a seamful digital phenotyping system that integrates users into the ML pipeline of passive sensing to support personalized and ethical mental healthcare. The system invites users to reflect on their sensed data (e.g., sleep, phone usage, activity), review AI-generated depression estimates, and modify model parameters—placing them in the loop.
This project aimed to reimagine digital phenotyping as a transparent, user-configurable experience, balancing the clinical power of wearables and smartphones with patient agency and trust.
My Contribution
As a UX Researcher and Designer, I:
Led the end-to-end research process, from study design to synthesis.
Designed and wireframed the DYMOND dashboard, mapping user stories and sitemaps.
Created data visualizations and journaling interfaces to support model transparency.
Delivered weekly insights to researchers and software engineers to guide development.
Conducted longitudinal in-the-wild field studies with Garmin wearables.
Moderated bi-weekly participant check-ins and co-design sessions for 20+ participants
Conducted competitive analysis with health AI experts (5+ analyses)
Read and synthesized 5+ research papers on depression monitoring and HCI
"I think I realized through Pearl and if I were to design it again might design it differently is that most everyday people don't know where to start with really Advanced statistical visualizations."
Matthew jorke, PhD Student at Stanford University
competitive research
Research Goals
Identify usability and trust barriers in passive sensing for depression.
Understand how users interpret, accept, or reject AI-driven mental health inferences.
Explore how participants want to configure wearables + ML pipelines.
Methods
Identify usability and trust barriers in passive sensing for depression.
Understand how users interpret, accept, or reject AI-driven mental health inferences.
Explore how participants want to configure wearables + ML pipelines.
Prototype human-in-the-loop workflows that enable transparency and control.
Insights
Trust Hinges on Transparency: Users trusted DYMOND more when they understood its reasoning.
After conducting 40+ hours of interviews, here are the top insights we've gained.
Are Valuable: Mismatches prompted useful reflection and system feedback.
One Size Doesn’t Fit All: Configuration motivations varied (privacy, reliability, utility).
Wearable Context Matters: Auditability increased users' comfort with passive data.
Feedback Loops Build Engagement: Users felt more respected and in control.
Outcome & Impact
Retention: 100% of completed participants actively engaged with configuration tools.
Design Impact: Informed redesigns of dashboard and feedback loops.
Research Impact: Presented at CHI 2025 Workshop; shaping ethical digital phenotyping frameworks.
More Projects
Designed with <3 by Olivia Wang