Our Projects
SafeRoad AI
Today, dangerous places are usually identified only after people have died there. SafeRoad AI aims to change that by combining evidence that currently sits in separate places (police crash records, hospital data, vehicle speeds, road imagery and traffic cameras) and analyzing it continuously. Six specialist agents each handle one kind of evidence. Three supervising agents coordinate them, check their conclusions, and record every step.
The agents never act on the road and never penalize anyone. They advise; authorized officials decide.
Rovanima AI
Rovanima AI is an AI agent framework that delegates daily back-office operations of tourism enterprises to a team of specialized agents while keeping a human coordinator in charge.
Small and medium-sized tourism enterprises (SMEs) are the backbone of winter tourism in Finnish Lapland, yet their daily back-office operations still depend on manual, email-driven coordination that limits how far they can grow. Rovanima does it with a team of specialized agents while keeping a human in charge: the coordinator requests the work, reviews it, and approves it, and nothing is committed without that approval.
Pipeline decisions come from a consortium of fine-tuned large language models (LLMs) whose outputs a reasoning LLM synthesizes. The design emphasizes responsible and explainable AI: model disagreement is escalated to the coordinator, outputs are traceable through typed records and an audit log, and the pipeline’s models run locally to keep guest data with the operator.
We implemented a prototype and evaluated it on simulated data for a Lapland winter tourism scenario.
