TG.SAG2.2 - fib Large Language Model

The fib has accumulated an unparalleled body of technical knowledge over decades, including design codes, guidelines, state-of-the-art reports, scientific publications, proceedings, and technical documents. While this knowledge forms the foundation of best practice in structural concrete engineering worldwide, its growing volume and complexity increasingly challenge efficient access, interpretation, and practical application.

The main objectives are to:

• Define the functional objectives of the fib knowledge-based digital framework, such as controlled knowledge retrieval, contextual interpretation of existing fib documents, decision-support assistance, design guidance, and recommendations aligned with best practices.
• Establish the scope and limitations of the framework’s application in structural concrete engineering, clearly distinguishing informational support from engineering responsibility and final decision-making.
• Curate, structure, and maintain a high-quality, authoritative knowledge base composed exclusively of fib resources, ensuring accuracy, consistency, traceability, and representation of diverse perspectives (e.g. structural engineering, materials, academia, and practice).
• Develop an LLM-based AI-supported system architecture optimized for domainspecific reasoning in structural concrete engineering, incorporating knowledge representation strategies that reflect engineering logic and terminology.
• Implement transparent and explainable AI mechanisms, allowing users to understand the origin, rationale, and limitations of AI-generated responses.
• Validate the framework through rigorous testing, expert review, and user acceptance testing (UAT) to ensure reliability, relevance, and alignment with fib’s objectives.
• Address legal, ethical, security, and data protection requirements, ensuring compliance with applicable regulations (e.g. GDPR) and alignment with fib ethical guidelines.
• Design processes for deployment, user support, performance monitoring, continuous improvement, and long-term adaptability of the system.

Convener
Sylvia Kessler

Co-Convener
Giulio Mariniello

View all members

get in touch

access network

documents

TG.SAG2.3 - Data Foundations and Ontologies

The successful and responsible application of artificial intelligence in structural concrete engineering depends fundamentally on the availability of structured, high-quality, and semantically consistent data. While the fib possesses a rich and diverse body of technical knowledge, spanning materials, design, durability, construction, monitoring, and lifecycle performance, this knowledge is currently distributed across documents, databases, and models that are not uniformly structured or semantically aligned.

The main objectives are to:

• Identify and classify key data domains within fib that are most relevant for AI applications, such as materials properties, structural performance, durability, environmental exposure, construction processes, degradation phenomena relevant to existing structures and structural health monitoring.
• Review existing fib documents, databases, models, and terminologies to identify overlaps, inconsistencies, gaps, and opportunities for harmonization.
• Develop shared ontologies and data models that define essential entities, relationships, and metadata in a machine-readable and consistent manner.
• Align fib ontologies with relevant international standards and initiatives (e.g. ISO, CEN, buildingSMART) to promote compatibility beyond fib.
• Establish interoperability frameworks that enable structured data exchange between fib databases, digital tools, AI workflows, and external systems.
• Support ontology-based integration into AI applications, enabling automated reasoning, information retrieval, and cross-domain analysis.
• Define a governance concept for maintaining, updating, and extending shared ontologies over time.
• Facilitate capacity building and alignment across fib commissions and task groups to promote consistent terminology and data modeling practices.

Convener
Roman Wan-Wendner

Co-Convener
Agnieszka Jedrzejewska

View all members

get in touch

access network

documents