LLMs

Pointing at a loss function is not a valid substitute for an actual engagement with the behaviors and internal representations a model has developed.

A multi-agent AI system for Entlastungsdienst Bern that assesses care workers’ competencies through conversation, creates personalised learning paths, and enables interactive training. Semi-finalist, GenAI Awards 2026.

An on-premise and Swiss-cloud platform for running open-source LLMs, giving businesses full control over their data. Deployed across multiple client industries.

An experiment letting LLMs converse freely with each other to explore emergent trajectories from different model pairings. Features qualitative and quantitative analysis of the resulting dialogues, with tools to search and inspect conversations.