EVA
A voice-first AI companion for elderly users. Press a button, speak. That's the entire interface. Now a working product, in pilot.
"The problem isn't their ability. The problem is that technology isn't designed for them."
Elderly users are systematically excluded from modern technology. Not for lack of ability, but because technology isn't designed for them: smartphones too complex, voice assistants tuned for speed, not patience. One confusing moment, one unexpected popup, and trust is gone for good. I saw this first-hand, and couldn't leave it alone.
One interaction model: press a button, speak. From there, EVA handles the rest.
- + Reminders and calendar management via natural voice
- + Answers questions in plain, patient language
- + Initiates and manages calls
- + Provides daily context ('what do I have tomorrow?')
- + No apps, no screens, no learning curve
Not a deck, not a prototype. A real product, built end to end and running today.
- Mobile app
- Expo / React Native app: one big button, real voice conversation, large-type reminders and a settings screen built for elderly hands.
- Voice pipeline
- Vapi orchestrating LLM + speech-to-text (Deepgram) + text-to-speech (ElevenLabs), tuned for patience over speed, with a Spanish-accent voice.
- Backend
- Supabase — Postgres, row-level security and Edge Functions wiring the assistant to real data and a family panel.
- Memory
- EVA remembers the person between conversations — private and on-device, it never leaves the phone.
- Real-time info
- Live web search (Tavily) so EVA can talk about today's news, weather and sport, not a frozen training cut-off.
A working MVP in pilot validation: a real mobile app talking with a real voice. The open question now isn't whether the tech works — it does — it's the human one: does a real elderly person understand it, trust it, and keep coming back? That's what the pilot is for.
See EVA in action.
The landing walks through what EVA does, with real demo clips.
Open the live demo