• Funding : CRIL
  • Start year :
  • 2026

This thesis proposal investigates the use of Large Language Models as agents in scalable, adaptive, and persuasive multi-agent environments. While recent work has shown that LLM-based agents can support task decomposition, collaborative reasoning, and strategic interaction, current frameworks remain limited when deployed in large-scale, dynamic, and unstructured settings. In particular, they face challenges related to coordination among many agents, conflict resolution, belief revision, long-term memory, and robust reasoning under changing conditions. The proposed research aims to address these limitations by developing gaming-inspired algorithms for communication and role allocation, belief revision mechanisms for dynamic environments, hybrid reasoning methods combining symbolic AI and reinforcement learning, and memory-augmented architectures for extended goal-oriented reasoning. The expected outcome is a new generation of multi-agent LLM systems that are more scalable, adaptable, and capable of complex reasoning and collaborative problem solving. This work has potential applications in strategic decision-making, persuasive environments, and robust AI-assisted collective intelligence.