How ONESTRUCTION built the Ishigaki-IDS foundation model with AWS GenAIIC
Domain-specific foundation model in construction: how ONESTRUCTION used synthetic data and three-stage training to solve BIM's data scarcity problem.
Why it matters
A practitioner building or evaluating domain models in data-scarce industries gains a concrete blueprint for synthetic data + training pipeline design. Enthusiasts tracking frontier labs' playbooks see a replicable pattern: AWS GenAIIC advisory + EC2 training + verifiable rewards.
The key facts
12 to knowModel: Ishigaki-IDS (construction and BIM specialized)
Partner: ONESTRUCTION with AWS Generative AI Innovation Center (GenAIIC)
Training approach: three-stage pipeline
Data strategy: synthetic data to address scarcity
Infrastructure: Amazon EC2
Evaluation: verifiable rewards (RLHF-adjacent)
Domain: construction/Building Information Modeling (BIM)
Model: Ishigaki-IDS, specialized for construction and BIM workflows
Technique: synthetic data + three-stage training pipeline + verifiable rewards
Partner: AWS Generative AI Innovation Center (technical advisory)
Problem addressed: data scarcity in construction domain
Published as AWS case study (Aug 11, 2026)
Go to the source
AWS Machine Learning Blogaws.amazon.com
Publisher excerpt: 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 verifiable rewards on…