Postdoctoral Fellow in Norm-Constrained Multi-Agent AI for Child-Safe Systems

Your role

AI4KIDS addresses the critical need for child-centric safe AI by developing a norm-first Belief–Desire–Intention (BDI) architecture where generative models (LLMs) are constrained by machine-readable child-protection policies to ensure purposeful, legally compliant, explainable and auditable AI behaviour. The research advances three main technical fronts:

• Embedding explicit norms into the BDI cycle, compiled into enforceable policy-as-code for real-time safety checks
• Parameter-efficient fine-tuning of LLMs for age-appropriate, pedagogically meaningful dialogue
• High-performance computing pipelines on Luxembourg’s MeluXina supercomputer for large-scale simulation and adversarial testing

The project also includes industrial validation with social robotics platforms (e.g., QTrobot) for deployment in educational and special-needs contexts, marrying computational law, symbolic AI, and large-scale evalua...