Our research focuses on evaluating the evolution of advanced AI agents — based on LLMs and other machine learning (ML) technologies — across a range of agentic frameworks, from open-source platforms to industrial-grade systems.
We aim to understand how agent societies evolve, adapt, and collaborate when powered by modern AI — and how these evolving systems can transform traditional business software workflows into more adaptive, intelligent, and resilient processes.
Core Research Areas
Evolving Multi-Agent Systems with Advanced AI Agents
We investigate how agents built on large language models (LLMs) and other machine learning techniques can develop cooperative, competitive, and adaptive behaviors, particularly in real-world industrial applications.
Our work explores how agent societies can replace or augment traditional rule-based and process-driven software, creating new opportunities for dynamic, autonomous, and scalable solutions.
Framework Evaluation and Benchmarking
We systematically evaluate open-source and industrial-strength agentic frameworks, assessing their ability to support evolving AI agents based on:
- Scalability and system throughput
- Communication flexibility and semantic richness
- Autonomous evolution and learning capabilities
- Integration into existing enterprise IT ecosystems
- Robustness and reliability under business-critical conditions
Emergent Behavior and System Dynamics
We study the emergence of:
- Cooperation, negotiation, and decentralized problem-solving
- Specialization and dynamic role adaptation among agents
- System resilience to failures and environmental perturbations
- Efficiency and innovation beyond static, hard-coded business logic
Particularly, we examine how advanced AI cognition at the agent level fosters these emergent properties at the system level.
Ongoing Projects and Future Directions
Comparative Evaluation Across Frameworks
We are developing a standardized evaluation suite featuring industrially relevant benchmarks such as:
- Logistics and supply chain optimization
- Smart grid management and energy trading
- Financial markets modeling and fraud detection
- Autonomous service orchestration for complex business workflows
Framework Adaptation and Enhancement
Based on experimental insights, we aim to propose improvements to agentic frameworks to better support:
- Evolutionary agent development
- Seamless multi-agent coordination
- Secure, efficient communication in large-scale deployments
Industrial Collaborations
We are actively collaborating with partners to apply evolved multi-agent systems to:
- Intelligent supply chain networks
- Financial modeling and risk management
- Smart manufacturing and production systems
- Municipal governance and citizen service automation
Vision
As industries move beyond rigid, rule-based automation, multi-agent systems powered by advanced AI agents — built on LLMs and machine learning — offer a transformative path forward.
Our research bridges the gap between traditional enterprise software and next-generation adaptive AI, enabling systems that are more flexible, resilient, and intelligent than ever before.
If you are interested in collaborating, partnering, or learning more, please feel free to reach out.