AI Health Guardian
Health videos and messages spread faster than older adults can check them. This service makes checking as easy as sharing.
Open KakaoTalk channelView code
The problem
Of the 100 most-viewed YouTube videos on senior health, 42 were made with generative AI, and 24 of those impersonated experts (Korea Health Promotion Institute). Many older adults forward such videos to family and friends without doubting them.
The approach
There is no new app to install. Older adults share a link or message to a KakaoTalk channel. The service extracts the claims, searches papers through Liner, checks Korean food-safety data, and returns a result card they can pass on. The scope is food and supplements. It never diagnoses.
How it is used
Four possible answers
- Trustworthy
- Be careful
- Warning
- Unknown
A rule table makes the final call, so the same evidence always gets the same answer. When the evidence is weak, the service says it does not know.
My role
I led the five-person team and did all of the planning and development myself. Teammates supported the volunteer sessions and the final presentation at the ideathon.
Results so far
- Key-claim recall rose from 40% to about 80% on 45 real links after adding an LLM extractor.
- Paper stance classification reached 89% accuracy on 383 pairs.
- A neurology professor reviewed 20 cases and found the verdicts broadly appropriate.
- At the first volunteer session, 20 of 34 surveyed older adults said they would use the service.
- Won the Chairperson’s Award at the Gyeonggi Volunteer Center Ideathon, placing 7th of 15 finalists from 101 teams.
Built with
Python, Kakao Chatbot, Liner Academic Search and Model APIs, Korean food-safety Open API, Docker, Fly.io