Towards Automated Evaluation of Multi-Turn Agentic AI Systems

• Unité de recherche: INNOSUISSE
• Numéro de projet: 129.524 INNO-ICT
• Titre du projet anglais: Towards Automated Evaluation of Multi-Turn Agentic AI Systems

Mots-clé (Anglais)

ICT, Artificial/local/extended/networked intelligence

Résumé des résultats (Abstract) (Anglais)

We propose an LLM-based framework for automated evaluation of multi-turn Agentic AI systems. By simulating diverse user interactions and assessing sub-task performance, it allows for an innovative approach for systematic testing, which reduces the reliance on expensive manual annotation.