Project Summary: 

This research develops an AI-powered decision-intelligence framework for reducing carbon across the lifecycle of façade systems. Façades are treated as systems in which geometry, materials, cost, constructability, compliance, maintenance, replacement, and end-of-life decisions affect one another. 

The research addresses a major limitation of current AEC workflows: BIM, life-cycle assessment, cost planning, regulatory analysis, and construction data remain fragmented across separate tools, files, stakeholders, and project stages. Although these systems exchange information, they commonly do not preserve the relationships between components or the reasoning behind design decisions. 

The proposed framework translates detailed façade BIM models into machine-readable graph data models. Façade components become nodes connected via spatial, semantic dependency relationships. A specialised façade ontology defines component identities, hierarchies, geometric relationships, attributes, constraints, and links to supporting knowledge. 

The framework is organised through three connected memory layers: 

1. Structural Memory: Translating BIM Models with high LOD to Graph Data Models that include relational data between facade elements, so AI becomes capable of systems-of-systems understanding. 

2. Reference Memory: reusable knowledge concerning materials, carbon, costs, façade typologies, regulations, construction, maintenance, and replacement. 

3. Decision Memory: a traceable record of alternatives, assumptions, evidence, agent interactions, human decision-making, and selected decisions. 

A planner agent coordinates four specialised agents: Material and Carbon, Cost, Lifecycle and Replacement, and building codes. These agents combine deterministic calculations, domain datasets, APIs, GraphRAG, and large language models to evaluate design alternatives. Their findings are compared within a shared reasoning process, allowing the consequences of a component-level change to be traced across carbon, cost, constructability, compliance, and lifecycle performance. 

Graph Neural Networks provide an additional analytical layer. As consistent project datasets become available, GNNs can recognise recurring relationship patterns, detect carbon-intensive configurations, and prioritise components for further investigation. They complement the ontology-based reasoning, domain analysis, and professional judgement. 

In this project, architects, engineers, cost planners, sustainability consultants, manufacturers, and contractors can inspect evidence, adjust assumptions, compare alternatives, and approve or reject recommendations. Their decisions are retained as reusable project and organisational knowledge. 

The framework uses graph databases, JSON/JSON-LD, and API-based queries to exchange information without repeatedly transferring complete 3D models. It can connect to environmental product declarations (EPD), EC3, Material2050, NABERS, regulatory sources, and industry cost and time datasets. 

The research delivers an integrated environment in which geometry, evidence, AI agents, analytical methods, and human expertise operate in a shared representation of the façade. Its purpose is not only to calculate lower-carbon outcomes, but to make the way toward those outcomes relational, transparent, traceable, and reusable across AEC projects.