Multi-Agent System for Project Risk Management
Leverages the Mixture of Experts (MoE) concept to build an AI-powered multi-agent system for construction project risk management.
This study is funded by Institution of Engineers Singapore (IES). The contributions of experts are acknowledged and greatly appreciated.
Construction projects globally face substantial challenges with schedule delays, budget overruns, quality deficiencies, and safety incidents
Our approach
AI technologies like LLMs and machine learning show promise in improving risk management, but no single model can cover all specialised domains effectively. Inspired by the Mixture of Experts (MoE) architecture, we propose a multi-agent system where each AI agent acts as a domain expert with its own structured knowledge base.
Our solution focuses on:
- Capturing & organising lessons learned
- Structuring risk data for easy access
- Delivering actionable insights for better decisions
- Optimising knowledge for continuous improvement
Overview of our multi-agent system for project risk management
Comprehensive project risk management knowledge database
(1) Building a rich knowledge graph from diverse sources (standards, cases, literatures)
(2) Systematically identifying critical project risk factors and their causal relationships
(3) Ensuring knowledge is structured, reusable, and easily accessible for practical decision-making
MoE-based multi-agent architecture design
(1) Designing specialised AI agents that act as domain experts
(2) Enabling agents to perceive, reason, act, and learn collaboratively to improve risk management outcomes
Publications
