Background
Most digital solutions for elderly people are screens with buttons. But many elderly struggle with screens — they can talk, they can listen, they just want someone to chat with. That was the starting point for Diana.
I built Diana as a proof of concept for a partner working in elderly care. The idea was simple: an AI that lives on a tablet, speaks Danish, and remembers who it’s talking to. No buttons, no menus — just a conversation.
What I built
Diana is a React app that uses Google’s Gemini Live API for real-time voice conversation. The care staff’s instructions to Diana were written with help from Claude. The user speaks, Diana listens and responds — with a warm, patient voice in Danish.
Tech stack
- React + TypeScript frontend optimized for tablets
- Gemini Live API for real-time voice AI with Danish language support
- Express for secure token handling (API key never exposed)
- Zustand for persistable memory — Diana remembers details between conversations
- Docker + nginx for deployment
Care staff can customize Diana’s personality and knowledge about each resident through settings. Diana never gives medical advice, never corrects memories, and speaks in short, clear sentences.
Results
Diana is a proof of concept, but it demonstrates something important: voice AI is good enough to be a real interface for people who can’t use screens. Not as a gadget, but as company. That’s the kind of AI I find interesting — technology that does something for someone who otherwise gets nothing.
