BIM_LLM_code_agent

About

BIM (Building Information Modeling) files in the IFC format are large, semantically dense structures that general-purpose coding tools handle poorly, so this research agent targets the AEC industry specifically. It answers natural-language questions about building models by generating and running Python code against ifcopenshell, supported by a LangChain multi-agent system with memory layers, FAISS vector stores, and a code knowledge base accessed through both Vector RAG and Graph RAG. It runs OpenAI models alongside local Ollama models (codegemma, qwen2.5-coder, llama3, llama3), and a Streamlit web UI makes it approachable for AEC professionals who do not write code. Outputs include extracted data tables and 2D/3D Plotly visualizations. The project is a research artifact tied to published academic work, actively iterated by its author with an English/Korean bilingual README, and used by BIM researchers and AEC practitioners exploring LLM-driven IFC analysis.