University of Bath Develops AI Tool To Predict Building Emissions From Simple Text

Researchers at the University of Bath have developed a first-of-its-kind artificial intelligence tool that can estimate a building’s carbon footprint from a simple conversational description. The system is designed to support architects at the earliest stages of design, when decisions have the greatest impact on long-term emissions.

The tool uses machine learning and natural language processing to predict embodied carbon, which refers to emissions associated with materials and construction across a building’s life cycle. By translating text descriptions into carbon estimates, it provides immediate feedback before detailed specifications are finalized.

 

Addressing Embodied Carbon Early

 

Buildings and infrastructure account for more than one third of carbon emissions in industrialised nations. While digital tools are widely used to optimise operational energy use such as heating and cooling, embodied carbon has historically been harder to quantify during early-stage design.

Traditional carbon accounting tools require detailed material breakdowns, precise quantities, and specialist engineering expertise. In contrast, the Bath-developed model allows architects to input broad descriptions that include materials, dimensions, and intended usage. The AI then generates a carbon estimate that evolves as more design detail is added.

Professor David Coley, Professor of Low Carbon Design at the University of Bath and one of the project’s authors, explained that the tool is not intended to provide final compliance calculations. Instead, it offers predictions that are sufficiently accurate to guide conversations and design decisions before specifications are locked in.

 

Read more: Zelestra Advances 27.5 MWdc Klevenow Solar Project in Germany

 

How The System Works

 

To address the limited availability of publicly accessible embodied carbon datasets, the research team trained the model on a synthetically generated dataset representing 150,000 buildings. This approach enabled proof-of-concept testing and model development. The system can later be retrained using higher-quality real-world data to enhance precision.

The natural language processing engine is designed to manage the ambiguity and variation typical of architectural descriptions. During testing, the system correctly identified key materials such as steel, concrete, and timber 80 percent of the time. It also demonstrated consistency by producing similar carbon predictions from different textual descriptions of the same building.

In industry trials involving 43 building professionals, the tool was integrated into live design workflows. In one example near Exeter in the United Kingdom, designers used the AI to assess the embodied carbon of a glass and masonry building throughout its development. Adjustments to insulation, wall assemblies, and glazing were made to reduce emissions, even before precise material quantities were known.

 

Explore OneStop ESG Marketplace: AI (Artificial Intelligence)

 

Beyond Carbon Metrics

 

In addition to estimating embodied carbon, the tool can suggest improvements related to natural lighting, thermal comfort, and acoustics. This broader environmental assessment could help embed sustainability considerations into architectural education and early-stage concept development without requiring specialist carbon expertise.

By providing instant feedback at the point where design flexibility is highest, the tool aims to reduce emissions before they are embedded into materials and structures. As more real-world embodied carbon data becomes available, the researchers plan to refine the model and explore expanded applications across the construction industry.

The development signals a shift toward integrating AI-driven sustainability insights directly into creative design processes, potentially accelerating progress toward net-zero targets in the built environment.