Last fall, I started working in the city and found myself in a situation neither of us expected. Suddenly, we found ourselves in daily development meetings with experts from one of the world’s largest software companies.
We were under intense time pressure to code an application unlike anything we had built before. The app needed to be up and running within three months so it could be tested at the Contest for New Music (UMK) at Nokia Arena.
The result of this project is known as Tampere VIP.
On the surface, VIP is a voice-controlled app that recommends restaurants, events, and activities in Tampere. There are already several city apps around the world that offer personalized recommendations. What we think makes Tampere VIP an exceptional innovation is the structure of the app.
Architectures based on AI agents open up new possibilities
Tampere VIP is based on an AI-based architecture, where different tasks are distributed among multiple complementary agents. One main agent receives user queries, parses them, and forwards them to the responsible subagents.
For example, when a user asks where to have lunch, one of the agents retrieves the restaurant data, another agent checks the event, and a third agent records the user’s location. Each agent uses only predefined trusted data sources. The VIP application used data from Visit Tampere, which hosts the city of Tampere’s event calendar and tourism website. The answers were then combined into one easy-to-use package.
We built the application in such a way that if the agent and the datasets it uses change, the purpose of the application changes without having to rebuild the entire system. This allows the same technology platform to be used for completely different purposes. There’s no need to develop voice-enabled services from scratch every time.
