Location: New York, United States
Program: Software Development
Year: 2025
Abstract
The integration of artificial intelligence (AI) into Building Information Modeling (BIM) workflows offers transformative potential for addressing global construction demands and sustainability challenges. This study presents a prototype system that combines geospatial data, large language models (LLMs), and BIM automation to streamline architectural design processes. Key results demonstrate significant efficiency gains, enhanced accessibility for non-expert users, and actionable insights into environmental impacts. This study demonstrates the potential of generative AI to democratize BIM tools, optimize design workflows, and foster sustainable development, paving the way for future advancements in AI-driven architectural innovation.
Keywords—LLM, BIM, AI, generative design, sustainability, design automation
Introduction
According to the United Nations, the global population is expected to reach 9.8 billion by 2050. To accommodate this rapid growth, 13,000 new buildings will need to be constructed daily, presenting unprecedented opportunities for the architecture, engineering, and construction industries. However, this demand is most pressing in emerging economies across Africa, which often face inefficiencies, limited industry expertise, and constrained financial resources, exacerbating environmental impacts such as high carbon emissions from construction.
To address these challenges, BIM has become an important tool, providing a centralized platform to design, analyze, and manage built assets. However, BIM workflows remain complex, requiring significant expertise to bridge design intent with software execution. This gap hinders its widespread adoption in resource-constrained regions.
The advent of Generative AI like LLMs presents a new opportunity to simplify BIM. By converting natural language into actionable commands, LLMs reduce the technical barrier to BIM, enabling architects and developers to interact with BIM tools intuitively. This study explores the integration of LLMs with BIM to efficient workflows and enhance accessibility, to meet global construction demand sustainably.
Prototype
In this study, we developed a prototype in Autodesk Revit 2024 that processes key aspects of the architectural workflow, including site generation, zoning analysis, generative design, and sustainability evaluation.
The process begins with address-based site boundary generation, where users input a project site location as a street address. A geocoding API (we used NYC Zoning Map in this study) converts this input into geographical coordinates, which are used to retrieve polygonal boundary data in GeoJSON format through GIS platforms (OpenStreetMap). These data, transformed into Cartesian coordinates via the Proj.NET library, are visualized in Revit using the Revit API to generate site boundary models.
Next, the system conducts zoning code integration. Based on the inputted address, an LLM (ChatGPT 4.0) extracts information such as lot coverage, setbacks, floor area ratios (FAR), and height limits from municipal databases (NYC Zoning Resolution), and retrieves them to the system for ensuring the subsequent modeling process compliance with regulatory constraints. Concurrently, users input project descriptions using natural language, including total gross floor area (GFA) and desired building programs (e.g., residential, retail, or office).
The project description inputs, combined with the boundary data and zoning code, are compiled into structured prompts for generative design. The LLM analyzes these data and proposes building massing strategies, including footprint, orientation, and room layouts, to fit previously generated site boundaries. The outputs are formatted in XML for integration with Revit API.
Subsequently, Revit API transforms these outputs into detailed BIM elements. Tasks such as level creation, massing model generation, and room layout subdivision are executed, ensuring the design adheres to user requirements such as target GFA.
Following modeling, a sustainability analysis will evaluate the building’s material costs and carbon footprint. These values are calculated by LLM and visualized in the user interface to support early-stage design decisions. Users can adjust layouts, orientations, and building attributes via natural language commands processed by the LLM. Revit models are updated in real time, creating a feedback loop until the design satisfies user preferences.
After the model refinement, the system generates construction documents, including plans, sections, and 3D views. Users specify drawing requirements, which the LLM processes to command the Revit API, streamlining the documentation process and ensuring efficiency.
Video for the Workflow of the Tool.