Zum Inhalt springen

Venture · 2026

MAUTZI. Technology that listens.

A voice controlled assistance system against loneliness in care. Operated in natural language, with no technical knowledge required.

Status
Prototype, tested with residents
Role
Founding team · Product · User research
Duration
NBank founder scholarship, Jan to Oct 2026

The case in 60 seconds

Problem
Many older people feel alone. Care staff have little time for conversation.
Audience
Residents of nursing homes and the care staff around them.
Our contribution
Together with Phine from idea to prototype, supported by NBank.
Key decision
Speech only. Whoever can speak can use MAUTZI.
Status
Working prototype, tested on site with residents.
Main limitation
Early stage. No regular operation yet, so no reliable usage figures.

Starting point

Loneliness is one of the largest invisible problems in care. Many residents have very few conversations a day.

Digital offerings often fail with exactly this audience because they assume screens, menus and technical knowledge.

Research and insights

We spoke directly with residents and care staff in a nursing home.

The insight is simple. Interest is not the barrier, operation is. Everyone already knows how to speak.

What MAUTZI does

MAUTZI runs as a tablet app in kiosk mode. An animated cat embodies the assistant, everything is operated by speech.

One button starts the conversation, then voice activity detection takes over, no wake word needed. Plus appointments, weather, news, a media library, voice games like bingo and memory, and a drink reminder every two hours.

Technical implementation

Client server system with a server side persona. It keeps answers to one to three sentences, German only.

Two speech paths depending on the situation. Free conversation runs in realtime over OpenAI, the games use a grammar constrained recogniser with Vosk.

Test with residents

We tested the prototype on site with residents. That shows how people really speak to MAUTZI and where it still breaks down.

In parallel I modelled the service processes in my bachelor thesis and derived design requirements.

Learnings and next steps

Closeness to the people you build for beats any assumption made at a desk. Almost every good decision came from a conversation on site.

Next up are consistent voice operation, a reliable fallback and a longer trial in daily operation.

Tools and methods

  • Voice interface
  • OpenAI Realtime
  • Vosk
  • AI product development
  • User research

Key insight

Closeness to the people you build for beats any assumption made at a desk. The best interface is the one nobody has to learn.