Program Overview
A six-month collaborative research program with University of Washington Master of Science in Information Management (MSIM) students as part of their Capstone project. Each phase focuses on a distinct dimension of responsible AI deployment in mental health contexts — building toward a comprehensive evaluation of He@lio's effectiveness, safety, and ethical posture.
6
Research Phases
6
Months Duration
UW
MSIM Capstone 2026
MSIM Capstone Researchers
University of Washington
Amber Lee · Casey Frizzell · Jaspreet Bhamipuri · Ko-Ching Lu · Rebecca Ko · Saba Ziadlou · Tim Joo · Trudy Xia · Veronica Lee
Project Sponsors
Compassion8Innovation
Joydeep Hazra · Brian Ho
Supported by: Amazon · Bellevue School District
Phase Tracker
Program Phases
January – June 2026 · 6 Phases
Study of Mental Health Topics
Teams received AI fundamentals training from Google experts and conducted in-depth research into mental health topics affecting teens. The goal was to ground the research in real-world youth mental health challenges before engaging with He@lio's AI system.
Do No Harm
Establishing safety guardrails to ensure He@lio does not suggest, imply, or reinforce harmful behaviours. This phase evaluates the chatbot's responses across high-risk mental health scenarios including crisis situations, self-harm language, and harmful coping suggestions.
From Testing to Building
With a safer AI foundation in place, the teams shifted from safety evaluation to product strategy. He@lio's core algorithm underwent a refresh and a new round of testing to ensure it continues to meet safety standards before reaching any user.
Strategy Meets Story
Building on March's product proposals, the teams continued refining He@lio's product strategy while extending research into human-centered design principles. That work culminated in startup pitch decks and drafted investor narratives, translating months of research into a compelling product vision.
Topic to be announced
Final Outcomes & Recommendations
Phase 1 — Study of Mental Health Topics
Teams received AI fundamentals training from Google experts and conducted in-depth research into mental health topics affecting teens. The goal was to ground the research in real-world youth mental health challenges before engaging with He@lio's AI system.
- Received hands-on AI fundamentals training from Google experts, including prompt engineering best practices for large language models (LLMs).
- Conducted extensive research into teen mental health topics including depression, anxiety, bullying, eating disorders, and substance use.
- Studied how to craft effective prompts to generate reliable, safe, and useful AI responses in a mental health context.
- Combined AI training with mental health research to begin preparing He@lio to recommend content that is safe, evidence-informed, and appropriate for students.
Phase 2 — Do No Harm
Establishing safety guardrails to ensure He@lio does not suggest, imply, or reinforce harmful behaviours. This phase evaluates the chatbot's responses across high-risk mental health scenarios including crisis situations, self-harm language, and harmful coping suggestions.
- Reviewed He@lio's response corpus against established mental health safety frameworks including Safe Messaging Guidelines (AFSP) and Crisis Text Line protocols.
- Identified and flagged response patterns that could inadvertently minimise distress signals or provide ambiguous guidance in crisis contexts.
- Began iterative prompt and guardrail refinement in collaboration with the Compassion8Innovation team to harden response safety.
Phase 3 — From Testing to Building
With a safer AI foundation in place, the teams shifted from safety evaluation to product strategy. He@lio's core algorithm underwent a refresh and a new round of testing to ensure it continues to meet safety standards before reaching any user.
- Conducted structured market research to map the existing mental health tool landscape, identifying where He@lio has the opportunity to lead.
- Completed product management training to build shared fluency around users, roadmaps, and outcomes.
- Each researcher developed and presented an individual product proposal, marking a shift from prompt-testing to product strategy.
Phase 4 — Strategy Meets Story
Building on March's product proposals, the teams continued refining He@lio's product strategy while extending research into human-centered design principles. That work culminated in startup pitch decks and drafted investor narratives, translating months of research into a compelling product vision.
- Began preparing the final Capstone report for the University of Washington.
- He@lio received another backend update, with continued refinements to improve overall stability.
This is a living document — updated as each phase of the pilot program progresses.
A final outcomes report with recommendations will be published upon program completion in June 2026.
More Research
Mesh AI: Toward a Non-Linear Model of Human-Centered Intelligence
April 27, 2026
Most AI systems answer the question you asked. This piece asks whether a system can hold what you have not yet said. A design philosophy on context as a network rather than a queue, and on what understanding would actually require structurally, ethically, and over time.
Read ArticleQuantum Intuition: How AI Can Make Quantum Concepts Accessible to Everyone
April 15, 2026
Quantum computing gets talked about constantly and understood rarely. This article starts from a different angle. Rather than teaching the mathematics, can AI serve as a translator that helps anyone build quantum intuition through language, interaction, and play?
Read ArticleAirlume: Improving School Air Quality Through Service and Data Analysis
September 30, 2024
A white paper that examines how school air quality can be improved through integrated sensing, filtration, and dashboard analytics. The project frames indoor air health as an operational education issue, linking environmental monitoring to student well-being, attendance, and learning outcomes.
Read Full Paper