AI INTERFACES | AI NAVIGATION | VISION IMPAIREMENT | ACCESSIBILITY
WATCH OUT!
DESIGNED FOR GOOGLE DEVELOPER GROUPS ON CAMPUS 2025 KOREA X JAPAN SW·AI HACKATHON: THE BRIDGE - CLOUD, CONNECT, AND CONTRIBUTE
ROLE
AI Researcher & Interaction Designer
DESCRIPTION
In light of the recent forest fires in Korea and the frequent earthquakes in Japan, I suggested that our team could create an application that could help the visually impaired safely evacuate from natural disasters. During this project, I was in charge of creating the interaction design for our team.
TEAMMATES
Ryan Yuen (Waseda Univ.), Ryota Tetsuka (Waseda Univ.), Minkyu Baik (Korea Univ.), Jihyung Ahn (Korea Univ.)
AWARDS
Merit Award Winner (Top 3)
TIMELINE
May 17-18, 2025
RESEARCH METHODS
Desk Research
TARGET USERS
Visually impaired people who want to escape from a natural disaster
TOOLS
Figma, Google Gemini
PROJECT INTRODUCTION
“Watch Out!” is an AI-powered system that helps visually impaired people find safe paths during natural disasters. It uses a camera and smart software on
a phone to understand the surroundings, then gives clear spoken directions. A smartwatch adds gentle vibrations that guide which way to go, while a separate app lets family or caregivers share important health details and track the person’s safety.
PROBLEM STATEMENT
Visually Impaired People Struggle to Evacuate During Natural Disasters
Visually impaired individuals face significant and multifaceted barriers during natural disaster evacuations, stemming from inadequate accessibility services, insufficient audio cues, and hindered evacuation routes.
For both Korea and Japan, natural disasters like forest fires and earthquakes pose
a significant threat to visually impaired individuals.
Villarreal, M., & MacPherson-Krutsky, R. (2025). Barriers and best practices for inclusive emergency alerts and warnings. International Journal of Disaster Risk Reduction, 19, 123–134. https://doi.org/10.1016/j.ijdrr.2025.105581
Demiröz Yıldırım S. (2025). Inclusiveness of disaster management for persons with disabilities in Türkiye from stakeholders' perspective. Scientific reports, 15(1), 22779. https://doi.org/10.1038/s41598-025-05464-5
Nguyen-Trung, K., Thuy, T. T. T., Anh, N. P., Cong-Lem, N., Huyen, D. T., Diu, L. T., Giang, N. H., & Simon, M. (2025). Vulnerabilities of people with different types of disabilities in disasters: a rapid evidence review and qualitative research. Disasters, 49(3), e12686. https://doi.org/10.1111/disa.12686
Demiröz Yıldırım S. (2025). Inclusiveness of disaster management for persons with disabilities in Türkiye from stakeholders' perspective. Scientific reports, 15(1), 22779. https://doi.org/10.1038/s41598-025-05464-5
USER RESEARCH WITH GEMINI
Before designing our app, I was in charge of conducting a brief user research process using Google's Gemini to come up with user personas.
The Visually Impaired Father
Goals
-
Safely evacuating from a disaster.
Pain Points
-
It is hard to gain access to important information related to evacuating, because he lost his vision
-
Many alerts are visual, without accompanying visual cues.
COMPETITIVE BUSINESS MATRIX
_edited.png)
OUR IDEA
What if... we can provide an
effective disaster evacuation guide
for the visually impaired with AI?
KEY FEATURES
Voice Instructions as Directional Cues with Gemini
Automatically Alert Emergency Contacts
Smart Watch Compatibility
INFORMATION ARCHITECTURE FOR APP
We divided the application's target users into two groups: caregivers and the visually impaired. Caregivers will be able to add the information of the visually impaired, as well as get an automatic alert during a disaster. For the visually impaired, they will be able to talk to Gemini to ask for help, receiving voice instructions to evacuate.
INFORMATION ARCHITECTURE FOR SMART WATCH (TENTATIVE)
I also suggested the information architecture for a smart watch, which vibrates to tell users if they are going in the right direction or not. The prototype for the watch app was not created for this hackathon.
LO-FI PROTOTYPE
Watch Out! Keeping Your Safety in Sight, Always
FOR CAREGIVERS
.png)
Create an account for your loved ones
Add medical info to receive personalized medical help
We notify nearby medical staff of pre-existing medical conditions in case the user asks for medical help.
.png)
FOR THE VISUALLY IMPAIRED
Simply press wherever on screen for help
A red screen will appear after an emergency alert is issued. You can start describing your current situation, and Gemini will provide tailored evacuation guidelines accordingly.
SMART WATCH (APPLE WATCH) INTEGRATION
.png)
For tactile feedback,
add the app to smart watches
Apple watch screen design
In case sight and sound fade, guidance you can feel will still remain.
COLLABORATING WITH TEAMMATES
Impact
Inspired by recent natural disasters, including nationwide wildfire outbreaks in Korea and a magnitude 4.3 earthquake in Hokkaido, this project developed an AI-driven navigation system to help visually impaired individuals evacuate safely during emergencies. We aimed to integrate computer vision with Gemini AI to generate personalized, situation-aware audio guidance, addressing the lack of context in traditional audio alerts. By collecting user-specific data, such as age and health conditions, alongside live disaster parameters, the system enables tailored evacuation strategies. A smartwatch interface also provides pitch-based haptic signals to convey directional information without relying on vision or complex audio instructions. While currently restricted to earthquakes and facing limitations including internet dependency and potential LLM inaccuracies, we propose further strategies such as running local LLMs, implementing strict prompt engineering, and downloading basic information about building structures and terrains during internet access.
Overall, we brought up these ideas to advance assistive disaster navigation through an AI-driven approach that can be expanded to different disaster types, geographic areas, and other disability groups.
My Key Takeaways
-
Collaborating with teammates from an international background requires finding a common ground that everyone can relate to. My team spent a long time trying to figure out a topic we were all interested in. So, when I brought up the idea of disaster navigation, mentioning the frequent earthquakes in Japan and the catastrophic forest fires in Korea during the spring, my teammates could easily relate to the problem and be engaged in the project.
-
Scalability also matters for projects done in short durations. Another problem that my teammates and I had during the project was deciding on the scope. Though we had this big idea of creating a disaster navigation, we had trouble accessing and providing real-time data to users. Especially for forest fires, disaster navigation could be very tricky because the fire could create new obstacles in paths that the visually impaired could take. Furthermore, accessing data about where the fire started was also a challenge. Hence, we decided we would primarily work on earthquakes due to time constraints. I would definitely love to explore more on this project, and conduct user research on how AI could be used further to help users of other disability groups during disaster situations.