A student's AI prototype that asks how you feel before it tries to help

Problem

During a medical leave, Selah Key tracked and studied her own health numbers and read about what long-running stress does to the nervous system, the brain and the body. She noticed how many people live with those effects without knowing where they come from.

“I learned that many people interpret these experiences as personal failings rather than understanding what’s happening in their nervous system,” she told UW. “I wanted to create technology that helps people better understand themselves.”

In her ITU entry, the founder writes that the reactivity, withdrawal and impulsive behavior that chronic stress can cause are often mistaken for weakness or personality. Survivors of childhood adversity, domestic violence and relational trauma, she writes, are hit hardest and face some of the biggest barriers to care.

AI solution

Ordinary Days is designed as a phone app. It reads data a person already has in Apple Health or Oura: heart rate variability (the small changes in timing between heartbeats), heart rate, sleep, breathing and skin temperature. For the first 14 days it only learns, building a picture of that one person’s normal range rather than relying on population averages.

After that, it watches for moments when the body moves outside that range. A body signal alone does not trigger advice. The design has the app first ask, “Are you actually feeling stressed?” If the answer is yes, it asks about patterns the person may be noticing, and then, “What happened?” In the ITU entry, the founder adds that the user’s answer overrides the AI’s label, that prompts are written as questions, and that each one has a “Skip” or “Not now” option.

How it works

1. 01 → Learn For the first 14 days, the app only learns one person's normal range from their wearable data.
2. 02 → Ask When the body moves outside that range, it first asks: "Are you actually feeling stressed?"
3. 03 Suggest It learns which short body-based practice to suggest, and when. Every prompt can be skipped.

Three AI models sit behind those questions. One learns the person’s baseline. A second combines body signals with phone patterns (voice pitch and pauses, typing rhythm, time at home) and the person’s own journal keywords, then sorts the reaction into a type: Key’s diagram shows a revved-up “fight / flight” state and a shut-down “freeze” state. A third learns which practice to suggest and when, favoring those followed by a faster rebound in heart rate variability and good ratings from the user.

The suggestions are short and physical. For a revved-up state, the diagram lists wall pushes, wringing a towel, humming and walking in place. For a shut-down state, it lists orienting, a self-hug and cupping the face with the hands.

The ITU entry describes processing on the phone, with only encrypted model updates shared. Key’s own diagram also shows a “Send Data” step after a stress signal is detected, without saying where that data goes, and we found no published privacy policy. For now, it is fairer to say it is designed to keep raw health data on the phone.

Impact

Ordinary Days is at the concept stage. UW calls it “an end-to-end prototype”. We found no published user numbers, accuracy figures or outcomes, and no public app, website or waitlist. No study has shown whether the app helps anyone.

The practices draw on body-based trauma therapy. Among the studies the ITU entry lists is a 2017 randomized trial of Somatic Experiencing, a body-focused therapy, with 63 people with PTSD. The 33 assigned to therapy had 15 weekly sessions, and afterward they had fewer PTSD symptoms than the 30 on a waitlist. That trial tested a course of therapy sessions, so it does not show that short prompts from an app work. The entry also says the system follows the Polyvagal Theory, an account of the nervous system that other researchers dispute and its author continues to defend.

What is on record is recognition. In February 2026, Ordinary Days was one of two alternates named alongside 21 finalists in UW’s Hollomon Health Innovation Challenge, which drew 67 teams. In April, Key presented a poster at the HCDE Research Showcase in Seattle. On July 7 she presented the project at the Innovate for Impact session of the AI for Good Global Summit in Geneva, where, UW reports, it received a Best Use Case award through ITU’s Innovate for Impact initiative. ITU says it honored 59 projects at that session, chosen from nearly 400 entries, but its announcement does not list them all by name. UW also says the work was accepted to three workshops at CHI 2026, a conference on human-computer interaction.

“This was my first time presenting not to pitch an idea, but to share a solution that had been evaluated, recognized, and celebrated by experts from around the world,” Key said. As far as public sources show, that evaluation was expert review for the award, not a trial with users.

Ordinary Days is not a treatment and does not replace care from a doctor, therapist or counselor. If you are in crisis, please contact your local emergency number or a crisis service in your country.

People

UW describes Ordinary Days as Key’s independent research project, and she leads it as founder and lead researcher. She has more than seven years of industry experience in product, design and marketing, according to her AI for Good speaker profile. At UW she has also worked as a graduate researcher in research groups that include one on aviation mental health.

“HCDE gave me the flexibility to combine human-computer interaction, AI, clinical neuroscience, and design research into one project,” she said. According to UW, AI prototyping courses gave her the confidence to build the prototype. “I learned to think about the invisible architecture behind AI,” she said. “How the model learns, what data it uses, and how people interact with it are all part of designing a human-centered system.”

“It has strengthened my desire to see this work scale to a population level.”Selah Key, founder of Ordinary Days, to UW HCDE, 2026

Why it matters

Wearables can now measure a lot about the body, but the numbers do not say how a person feels or what happened to them. Ordinary Days is built to keep people in charge of that meaning: the AI is designed to check in first and to accept “not now” as an answer. Whether that approach helps people living with stress is the next question, and only careful testing can answer it.

What we could not confirm

• The latest public update we found is UW’s article from July 2026. We found no public app, website or waitlist as of October 2026.
• Only UW names a Best Use Case award. ITU’s announcement does not name Ordinary Days among the 59 projects it honored.
• The founder’s ITU entry says initial private pilots are under way. No details or results are public.
• We could not confirm what the “Send Data” step sends, or where.
• We found no public information on who else works on the project, who reviewed the practices, or any ethics approval for testing with people.