AI-Driven BIM Classification and Metadata Structuring for SketchUp Components

Functional Requirements

• Train and fine-tune an agentic AI workflow capable of automatically classifying and enriching SketchUp-native 3D components with IFC4.x-compliant metadata with at least 95% accuracy with indication of unpredictable components feedback
• Develop an approach that has flexibility in loading and adapting to different IFC schema for local compliance with BCA CorenetX (i.e. IFC-SG)
• Enable real-time classification and metadata injection directly within design workflows via modelling software plug-ins, starting with SketchUp Ruby/C API
• Web-based development for customer in-house digital content management platform, with these features:
  
  • Support batch processing of component libraries to streamline digital asset onboarding, with informed progress from the platform on estimation of remaining time, and aborting and resuming tasks
  • Able to deploy to an in-premise cloud platform via API (in order of priority: Microsoft Azure, Alibaba Cloud, and AWS)
  • Support both project-specific, and client-specific modes of content management
  • Support storing of metadata mapped to individual components

Technical Requirements

Core Requirements:

• Full compatibility with SketchUp (.skp files), with future extensibility to Revit (.rfa, .rvt), ArchiCAD, and Rhino
• RESTful API access to enable secure communication between design tools and the backend AI engine
• Modular agentic AI architecture encompassing classification, reasoning, validation, and formatting sub-workflows
• Compliance with IFC 4.x AND IFC-SG schemas, producing export-ready metadata
• Support enterprise deployment features, including Bring-Your-Own-Key (BYOK) capability for third-party API key-based processing, or in-premise infrastructure processing (e.g. gpt-oss, Gemma 3, Qwen-VL, etc.)‍

Additional Preferences:

• Real-time classification and metadata injection during component creation or import 
• Native integration with Trimble SketchUp AND/OR Trimble Connect for model federation and cross-disciplinary coordination 
• Plug-ins for additional BIM model authoring tools (in order of priority: Rhino, Blender, ArchiCAD, and Revit) to enable in-app access to AI functionality
• Transform disorganised or unstructured geometry into structured, federated digital assets ready for OpenBIM workflows

‍Expected Outcomes

• Reduction in early-stage BIM IFC model preparation time by 75-85% (typically equivalent to manpower cost savings of at least 60%)
• Improvement in consistency, efficiency and drastically reduce time to organising data in digital deliverables across teams

Deployment Environment and Constraints

Even for a modest architectural BIM model, such as a small residential or low-rise building, the number of components often exceeds 2,000 objects. In the case of M Moser’s typical workplace interior projects, this number routinely exceeds 5,000 components. The solution must be able to handle those large numbers. 

Also, the environment includes both in-house and external content with inconsistent standards. The solution must integrate with desktop modelling tools and support cloud-based collaboration workflows. It should work within project teams spread across regional offices with varying levels of BIM maturity and network connectivity.

M Moser will conduct the pilot within several ongoing workplace interior fit-out projects in Singapore. It should be able to run batch classification across >1,000 in-house and external components, validate all outputs through manual cross-check, and be benchmarked on classification accuracy, layer conformity, and time per component.

M Moser will provide:

• Curated Validation / Test Library: A verified dataset of 4,000 SketchUp components sourced from M Moser’s in-house library, 3D Warehouse, and project-specific content, spanning IfcFurniture, IfcLightingFixture, IfcElectricalAppliance, IfcDoor, IfcWall, and more. Each file includes benchmarked classification references for dataset validation and test.
• AI-Ready Infrastructure: A fully operational agentic pipeline, structured into four modular AI roles (Visual, Classification, Validation, and Formatting), is already integrated with the SketchUp Ruby/C API. This enables plug-and-play testing and seamless I/O for external teams.
• Real-World Projects: Multiple workplace projects in Singapore, Kuala Lumpur, Manila, Hong Kong, Shenzhen, and Guangzhou are available for pilot testing. These sites reflect realistic BIM conditions and workflows, allowing for robust use-case validation in both live and sandboxed environments.
• Domain Expertise and Advisory: M Moser’s internal GAIA (Generative AI Agency) global team and Singapore office consisting of over 100 professionals from various disciplines will provide constant feedback loops throughout the development stage. Additionally, the M Moser team includes advisors and practitioners familiar with IFC schema, ensuring the solution aligns with local and international IFC schema requirements.
• Integration Support: We will provide a sandbox enabled for the selected Solution Provider with unlimited storage and full access with Trimble Connect Workspace API Access for testing federated workflows and various real-life coordination and collaboration scenarios.

Throughout the pilot, M Moser’s BIM specialists, MEPF Engineers, Workplace Designers, Project Managers will participate in structured testing and review sessions, offering detailed insights into classification reliability, efficiency gains, and the tool’s fit within existing coordination workflows. Key performance indicators will include classification precision, average processing time per object, and interoperability with other BIM platforms, such as Trimble Connect, Graphisoft ArchiCAD, Autodesk Revit and IFC-compliant environments.‍

Commercialisation and Scaling

If validated through the pilot, the solution could be positioned as a scalable SaaS platform with tiered adoption models. Smaller design teams may benefit from free starter tools or pay-per-use APIs, while larger organisations could adopt enterprise licensing with in-premise deployment options.

The platform has potential to integrate with existing CDE providers such as Trimble, extending its reach across federated BIM workflows. With an estimated 500,000 BIM projects undertaken globally each year, the addressable market for AI-assisted classification and metadata enrichment is projected to be in the region of S$250 million annually.

The return on investment could be significant. By reducing BIM preparation time by 75–85% equivalent to manpower costs reduction of 60%, the solution has potential to deliver tangible productivity gains for project teams, while improving quality and consistency of digital deliverables.

Future scaling could focus on:

• Expanding support beyond SketchUp to Revit, Rhino, ArchiCAD, and other model authoring tools
• Partnering with regional and global stakeholders to align with local IFC schema (e.g. IFC-SG)
• Developing flexible licensing models suitable for both SMEs and multinational firms

By reducing manual BIM preparation, improving reliability, and enabling consistent digital deliverables, this approach may become a preferred pathway for accelerating OpenBIM adoption worldwide.