Automated compliance checking (ACC) evaluates BIM models against regulatory require- ments as well as domain-specific information needs, commonly defined through Employer’s Information Requirements (EIR) / Auftraggeber-Informationsanforderungen (AIA), Levels of Development (LOD), and Levels of Information Need (LOIN), all of which are becoming increasingly complex, time-consuming, and manual to manage. Ensuring that BIM models satisfy these requirements is essential for project quality, and they must be checked for compliance at multiple project stages. This thesis addresses this gap by proposing an LLM-driven compliance-checking framework that translates domain requirements and user questions into executable checks and orchestrates multiple reasoning backends, treating compliance as a multi-tool decision-making process rather than a single code-generation task. The system dynamically selects and chains three tool modalities: SQL queries for scalable property and attribute checks, Cypher graph queries for topology and containment reasoning, and IFCOpenShell scripts for geometry verification. At its core, the framework integrates a fine-tuned large language model, Devastral Small 2 24b, trained on domain- specific samples, including compliance prompts, cross-tool code-generation examples and patterns. This model is further tested using a tool-specific knowledge base and investigates various domain-adaptation techniques for code generation. The approach is evaluated on compliance tasks based on German regulatory prompts using IFC models with different schemas, reporting execution success and output-level metrics across attribute, relational, topology, and geometry checks, and analysing how task-specific adaptations influence text-to-code generation. The results indicate that multi-tool orchestration can be adopted for real-world compliance scenarios spanning attributes, topology, and geometry, and that this paradigm applies across different phases of domain-specific projects.     «

Automated compliance checking (ACC) evaluates BIM models against regulatory require- ments as well as domain-specific information needs, commonly defined through Employer’s Information Requirements (EIR) / Auftraggeber-Informationsanforderungen (AIA), Levels of Development (LOD), and Levels of Information Need (LOIN), all of which are becoming increasingly complex, time-consuming, and manual to manage. Ensuring that BIM models satisfy these requirements is essential for project quality, and they...     »