This research area focuses on hybrid intelligent systems combining Large Language Models (LLMs), autonomous AI agents, classical Multi-Agent Systems (MAS), symbolic reasoning, and neuro-symbolic supervision mechanisms. The research investigates how LLM-driven agents can be integrated with controlled and verifiable architectures to support safe, explainable, and adaptive decision-making in complex distributed environments.
Particular attention is devoted to Safe-by-Design hybrid architectures in which autonomous AI agents operate under symbolic supervision and workflow-aware behavioral control. The research explores how multi-step LLM reasoning processes can be represented as event logs, workflows, and logical structures enabling behavioral analysis, semantic drift detection, anomaly identification, and formal verification of AI-generated decisions.
An important direction concerns hierarchical MAS–LLM systems integrating context-aware reasoning, autonomous coordination, explainable decision-making, and adaptive supervisory control. The research also investigates symbolic guardrails, logic-aware prompting, attribution-based diagnostics, and self-correcting AI agents capable of improving reliability and reducing hallucination-related inconsistencies.
Special attention is given to workflow-level explainability and cooperative game-theoretic attribution methods, including Shapley-based analysis of contribution, criticality, and behavioral risk within autonomous AI workflows and distributed agent interactions.
Potential application areas include intelligent environments, IoT ecosystems, distributed autonomous systems, robotics, decentralized AI systems, and safety-critical operational infrastructures. Expected outcomes include new hybrid LLM–MAS architectures, supervised AI agents, workflow-aware neuro-symbolic frameworks, explainable autonomous systems, and advanced methods supporting trustworthy AI-driven decision-making in distributed intelligent environments.