911±ŹÁÏÍű

MSIM team designs guardrails for mental-health chatbot

By Mary Lynn Lyke Monday, June 1, 2026

You’re a teen, hurting and confused, certain no one in the world has ever felt so low. You’re too embarrassed to talk to friends about it or go to the school counselor, and your parents are 
 well 
 your parents. Therapist? No way. Who are you going to turn to?

An AI-powered chatbot now in development may provide teens some needed solace and advice. It’s called He@lio and it’s being vetted by two Capstone teams from the UW’s Information School: team Infomaniacs (Rebecca Ko, Amber Lee, Veronica Lee, Koching Lu and Saba Ziadlou) and team Insight Innovators (Jaspreet Bhamipuri, Casey Frizzell, Tim Joo and Trudy Xia).  

The Master of Science in Information Management students are working with sponsor Compassion8Innovation, a non-profit that researches and builds tools “where compassion meets technology.” Together, with funding from Amazon, the non-profit and iSchool teams are revising the chatbot pilot to ensure effectiveness and safety.

Row of five images of faces.
Team Infomaniacs, from left: Rebecca Ko, Amber Lee, Veronica Lee, Koching Lu and Saba Ziadlou

“The teams can design guardrails that implement ‘do no harm’ philosophy in software design and coding,” said Compassion8Innovation sponsor Joydeep Hazra, who praised the hard work of the MSIM students. “They will become great ambassadors of AI and benefit the industry with their newly acquired experience and skills.”

, developed with input from professional mental health experts, offers teens validation and calming advice. They may be stressed from cyber-bullying, suffering from eating disorders, or simply agitated after an argument with their “clueless” parents. He@lio’s response: “I’m sorry you’re dealing with that. Arguments can really stir up emotions. Try taking a few slow, deep breaths, or even go out for a short walk to clear your mind. Sometimes a little space and movement can help you feel more grounded and ready to tackle things again.”

He@lio also suggests teens practice goal-setting (tracking is built into the chatbot) and journal writing (prompts included).

If the teen is desperate — thinking of self-harm or of harming another — He@lio points them to hotlines where there is immediate crisis help and abundant resources. “This is heavy,” says He@lio. “We really want you to reach out.” The bot will flag severe cases. That’s when an actual human may step in. And that’s when a problem arises.

Row of four images of faces.
Team Insight Innovators, from left: Jaspreet Bhamipuri, Casey Frizzell, Tim Joo and Trudy Xia

Teens have to trust the chatbot before they will pour out their heart to it. They seek privacy. They don’t want parents and counselors involved. They’re reassured by He@lio that their information will not be shared with a third party. But if it’s a dangerous situation, it may be critical to round in adult help. How to weigh crisis care and privacy? “It’s one of the biggest barriers He@lio faces at this point,” said Casey Frizzell. “Privacy is the elephant in the room.”

Statistics show the need for more mental health tools for teens. One in three youths ages 12-17 experience mental, emotional, developmental or behavioral health challenges, according to the National Survey of Children’s Health. Anxiety is the most common condition, followed by depression. Yet many teens don’t have access to support systems. “It’s important for He@lio to have a low entry barrier,” said Veronica Lee. “Not all students have health insurance or money to go to therapy. We want to make sure every student can access some kind of mental health care and be guided by the right steps.”

Unfortunately, many distressed teens turn to popular tools not designed with their well-being in mind and not developed with professional therapist input. “It’s important when you look at the landscape of current AI tools — Claude, ChatGPT, Grok, Perplexity and others — to realize the tools are not tailored to teens, but are developed as catchalls that do nothing to help them,” Frizzell said. 

He@lio began its virtual life in 2021, during the pandemic, when members of the Bellevue Youth Link Board and Youth Council ideated the chatbot to address rising mental health challenges among young people. After learning computer science skills, they built the first version themselves, starting with Microsoft Virtual Agent and proceeding to a full standalone web application. Early iterations of He@lio were far from marketable. Responses were too unengaging and too generic: versions of “Oh that’s too bad. Go see a counselor” repeated again and again. “You’re not going to get a teen to interact with that,” Veronica Lee said.

The UW teams stepped in in January this year, starting with training in AI fundamentals from Google experts. They researched teens’ mental health challenges, established safety measures, revised and created prompts to mirror real-world conversations, and tested the tool using human-centered design principles. They refreshed the algorithm and went through more rounds of safety testing and refinements before they moved on to market research to study ways to roll out the product. 

Part of what attracted team members to the project was memories of living through their own vulnerable adolescence. “I was rebellious in my teenage years,” said Koching Lu. “My family did not know what to do with me. You had no chatbot to go to when you were crazy and stressed. He@lio would have been really beneficial.”

Inputting hundreds of mental health scenarios into the chatbot brought up past emotions for some team members, who found themselves reliving their own teenage traumas. “But that gives us some passion to put in the project,” Frizzell said. “How can we make this a little bit better for the next generation? How can we make that journey from teen to young adult a little bit smoother?”

Team Infomaniacs and Team Insight Innovators are proud of what they’ve accomplished to date. “In the end,” said Frizzell, “we’re doing something good. And not everyone in AI research can say that.”