AI Assisted BIM Query System

Abstract

This thesis presents a proof-of-concept system for querying Building Information Modeling (BIM) data in Tekla Structures using natural language. The system integrates a large language model (LLM) with a custom REST API built on top of the Tekla Structures Open API, enabling users to retrieve model information without needing technical expertise in Tekla Structures.

The implementation involved constructing a simple UI, exposing model metadata through REST API on top of Tekla Structures Open API, defining a system prompt which guides LLM, and extending the system with a containerized Python environment to support accurate numeric analysis. The approach was validated on a simplified Tekla model through a set of representative test queries. The results show that the LLM correctly translated user queries into function calls, applied reasoning over structured data, and returned accurate, human-readable responses. Limitations were observed in edge cases involving ambiguous semantic mappings and combined filters, suggesting directions for future improvement.

This work demonstrates the feasibility of combining LLMs with live BIM data access, providing a foundation for more intuitive, conversational interfaces in structural engineering workflows.