Inspiration
We were inspired by the growing challenge of dementia and the devastation it causes to both patients and families. We realized that while medical treatments are crucial, there's also a dire need for a tool that can maintain cognitive connections and stimulate recall.
What it does
By creating a detailed profile of the user's life history, relationships, and important memories it personalizes the experience of the patient. Furthermore, it features an intuitive voice interaction system that responds to natural conversation, in the voice of a loved one that was recreated using the ElevenLabs voice cloning llm. Through the use of context-aware AI, the model maintains meaningful dialogues while adapting to the user's cognitive state.
How we built it
The frontend comprised of Next.js, Three.js, tailwind, and WebSpeechAPI for voice interaction. The backend utilizes MongoDB for storing user profiles and conversation history and Next.js API routes for server-side logic. Cloudflare AI is used for the language model integration. Authentication is done through NextAuth.js with Google OAuth for secure user authentication.
Challenges we ran into
Some challenges we ran into
- Voice Recognition Optimization
- Reponsive 3D Visual Design
- LLM context preservation at an efficient rate ## Accomplishments that we're proud of
- Responsive and slick UI that incorporates 3D modeling
- The interface is accessible for older people who are more typically diagnosed with dementia.
- Well constructed LLM that maintains personality and context throughout conversation
- Implemented real time voice recognition and cloning. ## What we learned
- The importance of accessibility-based design
- The complexity of working with voice recognition
- How to handle large contexts in LLM
What's next for Reminisce
Improving upon our foundation by expanding interaction capabilities and improving the model to better assess and respond to the needs of the patient.
Built With
- cloudflare
- elevenlabs
- llm
- mongodb
- nextjs
- tailwindcss
- text-to-speech
- three.js
- typescript
Log in or sign up for Devpost to join the conversation.