ONESTRUCTION Builds Construction-Specific Foundation Model with AWS GenAIIC.

Koyo Hidaka· August 11, 2026 View original

Key takeaways

  • Domain-specific foundation models can be built even in data-scarce fields.
  • Synthetic data is a viable strategy for augmenting limited real-world datasets.
  • Multi-stage training pipelines enhance model specialization and performance.
  • Cloud partnerships and infrastructure are crucial for advanced AI development.

Who benefits

ConstructionArchitectureEngineeringManufacturingReal Estate

Summary

ONESTRUCTION, advised by AWS Generative AI Innovation Center, developed Ishigaki-IDS, a specialized foundation model for construction and BIM workflows, utilizing synthetic data and a three-stage training pipeline on Amazon EC2.

ONESTRUCTION has successfully developed Ishigaki-IDS, a new foundation model specifically tailored for the construction and Building Information Modeling (BIM) sectors. This achievement was made possible through technical guidance from the AWS Generative AI Innovation Center. The project highlights an innovative approach to model development in data-scarce domains. The development process involved a sophisticated three-stage training pipeline, leveraging synthetic data to overcome the typical lack of real-world data in specialized fields like construction. The model was trained and validated using verifiable rewards on Amazon EC2, demonstrating a robust methodology for creating domain-specific AI.

Why it matters

This case study provides a blueprint for developing specialized AI models in industries with limited data, showcasing how strategic partnerships and advanced training techniques can yield powerful, domain-specific solutions.

How to implement this in your domain

  1. 1Identify data-scarce domains within your industry suitable for specialized AI models.
  2. 2Explore partnerships with AI innovation centers or cloud providers for technical advisory.
  3. 3Investigate synthetic data generation techniques to augment limited real-world datasets.
  4. 4Design a multi-stage training pipeline to refine model performance for specific tasks.
  5. 5Utilize cloud infrastructure like Amazon EC2 for scalable and efficient model training.

Original post by Koyo Hidaka

"ONESTRUCTION, with technical advisory from the AWS Generative AI Innovation Center, built Ishigaki-IDS, a foundation model specialized for construction and BIM workflows. This architectural case study shows how they combined synthetic data, a three-stage training pipeline, and ve…"

View on X

Originally posted by Koyo Hidaka on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses