Product library and gold dataset for multi-division architectural takeoff - BC-1014

Project type: Research
Desired discipline(s): Engineering - civil, Engineering, Engineering - other, Architecture and design, Social Sciences & Humanities
Company: TakeoffBOT.ai
Project Length: Longer than 1 year
Preferred start date: As soon as possible.
Language requirement: English
Location(s): Victoria, BC, Canada
No. of positions: 1
Desired education level: CollegeUndergraduate/BachelorMaster's
Open to applicants registered at an institution outside of Canada: No

About the company: 

TakeoffBOT is a British Columbia-based software company developing AI-enabled takeoff tools for architectural millwork and adjacent fabricated architectural scopes. Its current platform helps estimators turn drawing sets into structured, reviewable scope using ML-assisted drawing understanding, linked navigation between plans, sections, and elevations, and exportable reporting and pricing outputs. The company has already established a practical workflow in Division 06 architectural woodwork, built around AWI-based standards, reviewable scope, and estimator-controlled outputs.

The next stage is to extend this capability into selected adjacent fabricated scopes in Divisions 05, 06, and 12 that rely on similar drawing interpretation and bidding workflows. This project addresses that need by developing a standards-based product ontology and a gold-standard labeled dataset that can be used to train, validate, and quality-check future models for North American construction practice.

Describe the project.: 

TakeoffBOT has developed an ML-assisted workflow for Division 06 architectural woodwork. This staged 24-month research project will create the standards-based product ontology, annotation rules, and gold-standard labeled drawing dataset required to extend that capability into selected adjacent fabricated architectural scopes in Divisions 05, 06, and 12. The primary research scope is North America, using AWI-based standards and Canadian and U.S. bidding practice as the starting point. In addition, the project will formally assess whether the resulting ontology, tagging framework, and labeled data approach can be transferred to other English-speaking markets, including the UK, Australia, and New Zealand, and what documented adjustments are required for selected EU contexts.

The research challenge is to translate heterogeneous information from plans, elevations, sections, details, schedules, and specifications into a consistent product classification and labeling framework suitable for machine learning training and quality assurance. The intern will review standards and drawing sets, conduct practitioner interviews with subcontractors, estimators, and architects, define taxonomy and annotation rules, and help build and validate labeled pilot datasets using both real and synthetic drawings.

The project will also compare regional differences in terminology, standards references, drawing conventions, product categories, and attribute requirements across the target markets.

The main outputs will be a reusable product library, annotation guide, pilot and expanded gold-standard datasets, benchmark protocols for future model comparison, and a documented set of market-specific adjustment guidelines for applying the framework in North America, the UK, Australia, New Zealand, and selected EU markets.

Required expertise/skills: 

The ideal candidate will have a background in engineering, architecture, or construction technology, with strong ability to read and interpret architectural drawing sets, including plans, elevations, sections, details, schedules, and specifications. Familiarity with CAD-based workflows, building documentation standards, and product classification is important.