Project Summary: 

AI-Driven Framework for Decarbonising Façade Systems 

This research introduces an AI-driven framework for decarbonising building façade systems using a graph-based representation of BIM data and a multi-agent architecture. The core contribution is based on translating building 3D BIM models into structured knowledge graphs, where each node represents a façade component enriched with semantic metadata which includes geometry, material properties, lifecycle cost, and regulatory compliance. These enriched graphs are considered as training input for Graph Neural Networks (GNNs), enabling pattern recognition and predictive design reasoning across large datasets of real-world projects. At the centre of the system is a planner-agent loop that orchestrates four specialised agents: Material, Cost, Lifecycle, and Regulation. Each agent contributes domain-specific intelligence either with Python or LLM-based workflows and adds semantic information to the initial graph related to the BIM Model. Regulatory reasoning is enhanced through RAG, ensuring grounded compliance logic during design evaluation. Once the training is done, the GNN highlights high-carbon or potential carbon components. The system starts dialogues among agents to propose low-carbon, cost-effective alternatives. A GraphQL interface that allows for dynamic feedback loops and human-in-the-loop participation, enabling stakeholders to explore trade-offs between carbon, cost, and compliance across multiple configurations. The framework is designed for integration with different datasets (e.g., EC3, WT Partnership’s cost and time databases) and is capable of being aligned with both early-stage design and later circularity scenarios. The novelty of this research is in the intersection of computational design, knowledge graphs, and agentic AI in the AEC domain. By acquiring Agentic-AI into façade system components and enabling real-time multi-objective informed decision-making, this research sets the foundation for scalable, data-informed decarbonisation strategies and intelligent decision-making. 

The image illustrates the journey from façade systems to graph neural networks and into agentic-AI decision-making. 

• On the left, we see architectural façades with material-intensive systems that are difficult to optimise for sustainability. 

• In the centre, these façades are abstracted into graphs: lightweight structures where each node represents a façade element and its relationships. 

• On the right, the AI engine becomes visible: a decision-making layer where agents, trained on real-world datasets, guide designers towards low-carbon alternatives. 

This represents how AI illuminates decision-making, identifying exactly where carbon-intensive components sit within a design and suggesting smarter substitutions that balance carbon, cost, and regulatory requirements. 

Features of the Agentic-AI Tool 

• Multi-Agent Orchestration: Material, Cost, Lifecycle, and Regulation agents working collaboratively. 

• GNN Analysis: Detects carbon-heavy design patterns across façade systems. 

• Human-in-the-Loop: Allows architects, engineers, and contractors to interact with AI insights, rather than replacing their judgement. 

• Lightweight Data Exchange: Uses JSON serialisation and GraphQL instead of heavy 3D models. 

• Industry Integration: Direct connections to EC3, Material 2050, and WT Partnership datasets. 

Contribution to Industry 

This research is not just about theoretical models. It directly addresses the biggest barrier in the AEC industry: fragmented decision-making across design, cost consultancy, and sustainability. By providing a shared, AI-ready data structure, it: 

• Bridges design and construction through interoperable knowledge graphs. 

• Accelerates low-carbon adoption by making trade-offs explicit and navigable. 

• Supports future policy frameworks by offering transparent, data-driven reasoning. 

In short, this framework provides the construction industry with a scalable, intelligent tool to transform how we design façades, not just to look good, but to perform sustainably across their entire lifecycle.