Voice-first AI memory companion for seniors with early-stage memory loss. Talk to it like a friend — it remembers what matters.
🎥 Demo: Watch on YouTube
Original Repo: Link
Built at the AI Partner Catalyst Hackathon 2025 by:
- Chimdinma Jason — Backend Lead (FastAPI architecture, RAG pipeline)
- Epaphras — AI Integration
- Eniola — Frontend Lead
I designed and built the backend architecture and RAG pipeline:
- Architected the FastAPI backend to handle concurrent voice requests with real-time LLM context injection
- Designed and implemented the Retrieval-Augmented Generation pipeline using Pinecone vector storage
- Integrated the custom ElevenLabs voice for low-latency speech response
- Built the persistent long-term memory system that recalls user details and daily tasks
- Backend: FastAPI (Python)
- AI/LLM: Gemini 2.0 Flash
- Vector DB: Pinecone (for RAG memory storage)
- Voice: ElevenLabs (custom voice synthesis)
- Frontend: React
User speaks → ElevenLabs transcription → FastAPI receives request → Pinecone retrieves relevant memory context → Gemini generates response with full context → ElevenLabs synthesizes voice reply → Streamed back to user.
Traditional memory aids (notes, apps, calendars) require fine motor control and consistent typing — exactly what early-stage memory loss makes difficult. Reminisce prioritizes natural voice interaction so users can stay independent longer.
cd backend
pip install -r requirements.txt
uvicorn main:app --reload
cd frontend
npm install
npm run dev
Required environment variables:
GEMINI_API_KEYPINECONE_API_KEYELEVENLABS_API_KEY