Skip to main content Skip to secondary navigation

Developing Virtual Assistants for the AEC-FM Industry

Main content start

Research Team

Our Motivation:

“BIM has become increasingly popular as an information management tool in the AEC-FM industry. However, efficiently searching for information from BIMs is still challenging, particularly for many non-tech-savvy practitioners. Therefore, this research aims to explore virtual assistant solutions to make BIMs more accessible by providing a natural language interface.”


 

Icon Contribution

Research Contribution

A dynamic prompt-based virtual assistant framework for BIM information search.

Icon Problem

Problem

Practical Problem

As BIM grows increasingly complex, managing and retrieving vast amounts of data from design, construction, and operation become more challenging. 

Current retrieval strategies demand users to possess thorough BIM technological expertise, including mastery of complex user interfaces, data structures, terminologies, and structured query languages. 

A lack of BIM proficiency is one of the major obstacles in harnessing the benefits of timely, abundant information in construction and operation.

Conceptual Problem

The genetic diversity of BIM user queries presents a major hurdle for creating virtual assistants with intuitive natural language interfaces. 

Current natural language processing (NLP) and machine learning (ML) approaches require substantial engineering efforts and extensive training data. 

Despite generative pre-trained transformer (GPT) technology offering promising prospects, its potential has yet to be explored in the context of BIM information search.

Icon Solution

Solution

This research introduces “BIMS-GPT”, a prompt-based virtual assistant framework for information search from BIMs. 

This framework integrates BIM and GPT technologies to automatically interpret users’ NL queries, construct structured queries, retrieve the requested information from the database, and deliver summarized NL responses and associated 3D visualization.

Icon added value

Added Value For The Industry

This research contributes to the advancement of effective and versatile virtual assistants for BIMs in the construction industry as it significantly enhances BIM accessibility while reducing the engineering and training data prerequisites for processing NL queries.

Icon Cooperation partner

Cooperation Partner

Autodesk, USA
 

 

Icon Timeline

 Timeline

DateActivityOutcome

Summer 2022

Research became awarded: Developing Virtual Assistants for the AEC-FM Industry 

 

Fall 2022

Literature review on existing approaches for BIM information search

 

Winter 2023

Framework development (virtual assistant prototyping)

 

Spring 2023

Framework evaluation (case study on a hospital building and experiment on a natural language query dataset for BIM)

 

Summer 2023

Submitted paper to the Journal of Automation in Construction

Publication: Dynamic prompt-based virtual assistant framework for BIM information search

Project Summary

(Provides you with a brief and clear summary of the insights and outcomes at the end of the funded year.)

This research introduces “BIMS-GPT”, a dynamic prompt-based virtual assistant framework to support natural language (NL)-based BIM search by integrating generative pre-trained transformer (GPT) technologies. 

We developed a novel dynamic prompt-based process to understand users’ NL queries, retrieve relevant information from BIM databases, and deliver NL responses with associated 3D visualization. 

In a case study, we demonstrated BIMS-GPT’s functionality through a virtual assistant prototype for a hospital building. Evaluated on a BIM query dataset, our approach achieved an accuracy rate of 99.5% for classifying NL queries with incorporating 2% of the data in prompts. 

Consequently, this research contributes to the development of effective, versatile virtual assistants for BIMs in the AEC-FM industry, as it significantly enhances BIM accessibility while reducing the engineering and training data prerequisites for building NL interfaces.

Contact Person

If you want to participate in the project please reach out to Junwen Zheng.