Inspiration

There are over 60 million dementia patients globally and the progression of the disease can very. Unfortunately, my grandmother is one of the 60 million affected by dementia and has been dealing with this for the past 6 years. I was very close to her growing up and recently it has become my mission to build something that helps these patients live their best possible lives despite having dementia.

What it does

The application helps users reminisce by walking them through their old photos and asking them questions about the times that they are reliving.

How we built it

I build the application over the last 3 weeks using an expo frontend, python backend, the gemini live framework for AI, and then Gemini AI models. The models receives the list of photos, the tags that describe what is in the photos, the users information, and the current conversation, and tries to get the user to talk about what the remember in each photo. If the user cannot remember, that is perfectly ok and the agent is happy to talk about what it knows based on the tags and image descriptions left by the caretaker.

Challenges we ran into

The biggest challenge that we ran in to was getting the audio playback from the agent to be smooth and consistent. Eventually we settled on send all of the audio chunks from the backend to the frontend over websocket and then having a queue that plays back the messages with a buffer so that it doesn't run out of messages to play to the user and then play nothing while it waits for the next message.

Accomplishments that we're proud of

I am proud of what I accomplished in the three week window. I think it is a solid project that lays the ground work for a mobile app that is going to help a lot of people in the future.

What we learned

I learned a lot about building mobile apps with the expo framework. This was my first time building with expo and I really enjoyed figuring out how everything works. This was also my first time using the google models and framework and I really enjoyed that too. They have a really good variety of different voices and the models are multi-modal which gives my the freedom to build anything I can think of with them.

What's next for Reminisce

Since the end of this build I have brought on another person to the team and we are working on expanding this into a big mobile app that will have a variety of different features that are all targeted to dementia patients and their caretakers. While this is just a rough prototype of the reminiscence feature, we are targeting a release of a complete app for the beginning of summer 2026. This application will provide dementia patients with relaxing games, training exercises, reminiscence therapy, and a general AI agent to help them with any questions, concerns or medical events throughout their daily lives. I am also happy to say that all of these AI enabled features will continue to be powered by the google gemini models and frameworks.

Built With

Share this project:

Updates