A Two-Phase AI-Driven Approach to Automated Construction Planning Using Small Language Models for Activity Sequencing and Missing Task Prediction
Journal
Proceedings of the International Symposium on Automation and Robotics in Construction
Start Page
964
End Page
971
ISSN
24135844
ISBN (of the container)
9780645832228
ISBN
9780645832228
Date Issued
2025
Author(s)
Abstract
This research presents a two-phase AI-driven approach to enhance automated construction planning and scheduling, addressing common challenges such as project delays and budget overruns in the Architecture, Engineering, and Construction (AEC) industry, often caused by inefficient project management. Traditional AI methods rely heavily on static programming and manually annotated datasets, leading to inefficiencies. The proposed methodology uses prompt engineering to systematically extract and structure information from historical project schedules, creating datasets for training domain-specific Small Language Models (SLMs). In the first phase, fine-tuning SLMs, particularly mistral-7B, achieved 83.69% accuracy, outperforming traditional approaches. In the second phase, another fine-tuned SLM identified activity patterns, optimized sequencing, and predicted missing tasks, achieving a mean similarity score of 81.43%. The results demonstrate that structured prompt engineering significantly improves model accuracy and efficiency, reduces manual efforts, and addresses computational inefficiencies. This scalable, secure, and efficient approach sets a new standard in automated construction planning.
Event(s)
42nd International Symposium on Automation and Robotics in Construction, ISARC 2025, Montreal, 28 July 2025 - 31 July 2025
Subjects
Activity Sequencing
Construction Planning and Scheduling
Domain-Specific Instruct Fine-Tuning
Prompt Engineering
Small Language Models
Publisher
International Association for Automation and Robotics in Construction (IAARC)
Type
conference paper