Post by Narendar Kumar

I am pleased to share our recent publication: "Trustworthy Causal Reasoning in LLM-Based Multi-Agent Smart Building Systems: A Multi-Level Evaluation Across Observational, Interventional, and Counterfactual Tasks" published by Elsevier | Scopus in Procedia Computer Science and presented at the 3rd International Workshop on Causality, Agents and Large Models (CALM 2026), Istanbul, Türkiye. This work investigates how Large Language Models perform causal reasoning in smart-building environments and whether multi-agent collaboration improves reliability in safety-critical decision-making. Through an evaluation across observational, interventional, and counterfactual reasoning tasks, we found that naïve multi-agent collaboration can sometimes reduce causal accuracy rather than improve it, highlighting the importance of structured coordination and trust-aware agent design. For me, this paper represents a small but meaningful contribution within my broader PhD research at the Laboratoire CIAD , Université de Technologie de Belfort-Montbéliard (UTBM). My PhD focuses on trust calibration in multi-agent systems for smart buildings, and this study provides an initial step toward understanding how trust, causal reasoning, and agent collaboration interact in intelligent building environments. I would like to sincerely thank my supervisors Prof. Vincent Hilaire, Prof. Stéphane GALLAND, and Dr. Yazan Mualla for their invaluable guidance, support, and contributions throughout this work. Their expertise played a fundamental role in shaping the research. Looking forward to continuing this research journey toward more trustworthy, explainable, and human-centered multi-agent systems. #MultiAgentSystems #TrustCalibration #SmartBuildings #LLM