ONESTRUCTION developed Ishigaki-IDS, a foundation model built specifically for construction and building information modeling (BIM) workflows. The project was carried out with technical advisory from the AWS Generative AI Innovation Center (GenAIIC).
According to the case study, the team addressed the challenge of limited domain data by combining synthetic data generation, a three-stage training pipeline, and verifiable rewards. The training was run on Amazon EC2.
Why it matters
Construction and BIM represent a data-scarce field, which makes building a specialized model difficult. This case study describes an architectural approach for developing a domain-specific foundation model despite that constraint.
Who should care
Organizations in construction and BIM, along with teams working on domain-specific foundation models in fields with limited training data, may find the described methods relevant.