Development of the AI-Powered Disaster Preparedness Concierge “Sonael”

2025-09-10

For the 2nd AI Agent Hackathon, we developed Sonael, an AI-powered Disaster Preparedness Concierge, with the theme of utilizing AI in the field of disaster prevention. The knowledge and technologies gained through this development will be applied to future AI-driven projects and service development.

Background

It is said that there is about a 70% probability of a major earthquake directly hitting the Tokyo metropolitan area within the next 30 years. Although this number raises awareness of risk, it often does not translate into concrete preparedness actions.
Sonael was created to address this issue, aiming to support disaster awareness that does not easily turn into action.

Overview and Features

Sonael is a disaster-preparedness support agent designed to respond to both objective risk information and subjective concerns.

Its main features include:

①Risk Analysis by Location
Analyzes four types of hazards—flood, landslide, earthquake, and tsunami—based on latitude and longitude, and quickly (⏱ approx. 5 seconds) provides a risk level of High / Medium / Low.

②Hearing Individual Concerns
In addition to family composition and housing conditions, it gathers emotional concerns such as “I’m worried about my pet during a disaster” or “What if I can’t sleep at an evacuation center?”

③Personalized Disaster Preparedness Report
Generates a customized report combining risk assessment with personal concerns. Recommended preparedness items are referenced from official municipal materials, and required quantities are automatically calculated based on family size and the number of days of stockpiling

④Chat-Based Support
Users can ask questions as if confiding in a friend. The AI suggests additional items or alternatives. With each conversation, their personalized disaster-preparedness item list grows.

Technical Architecture

Frontend & Backend: Ruby on Rails 7
AI Platform: Vertex AI (Gemini / Embedding Models)
Data: Tokyo Metropolitan Government disaster-preparedness guidebooks and related materials stored in Firestore for search and retrieval
Deployment: Cloud Run

User Test Feedback

Trial usage provided the following feedback:

“I realized concerns I had never thought about before, such as menstruation or staying warm in an evacuation center.”
“I felt reassured being able to check whether my preparedness was sufficient.”

At the same time, requests were raised for improvements such as “more localized data like evacuation shelter information” and “suggestions for in-home safety measures like furniture anchoring.”

Conclusion

In the 2nd AI Agent Hackathon, we took on the challenge of designing experiences that make disaster preparedness personal and actionable through the perspective of ‘facts × personal concerns.’
Although disaster preparedness is something that should logically be a top priority, the real challenge was how to connect it to action. We wrestled with UI and storytelling design, working tirelessly toward the goal of creating an AI that anyone can use and act upon.

The insights and technologies gained through this development will continue to be applied in future AI-powered projects and service development.