Natural Language Information Retrieval from BIM Models: An LLM-Based Multi-Agent System Approach
Open accessAbstract
While Building Information Models (BIM) effectively store building-related information, accessing it requires specialized software and expertise. Natural Language (NL) interfaces for BIM data retrieval can mitigate this challenge, but existing approaches are limited by rigid ontological frameworks or extensive pre-processing requirements. We present a Large Language Model-based agentic workflow that processes NL queries and automatically interacts with IFC-encoded BIM models without ontological or pre-processing constraints. In tests across architectural, structural, and MEP domains, our approach achieves 80% overall accuracy. We provide open access to IFC-Bench-v1, our evaluation dataset containing various queries, answers, and reference BIM models.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Benchmarks such as IFCBench evaluate independent question-answer pairs, usually assuming that each answer can be derived from the IFC model alone.
“Benchmarks such as IFCBench [9, 10] evaluate independent question-answer pairs, usually assuming that each answer can be derived from the IFC model alone.”
From IFCMemoryBench: Evaluating Long-Term Memory of LLM-Based Agents in BIM Information Retrieval (2026) · cites this paper as [10]
The IFC models are drawn from the public IFC-Bench.
“The IFC models are drawn from the public IFC-Bench [9, 10].”
From IFCMemoryBench: Evaluating Long-Term Memory of LLM-Based Agents in BIM Information Retrieval (2026) · cites this paper as [10]
BIM retrieval benchmarks such as IFC-Bench treat tasks as independent, stateless queries and do not exercise long-term memory.
“BIM retrieval benchmarks such as IFC-Bench treat tasks as independent, stateless queries and do not exercise long-term memory.”
From IFCMemoryBench: Evaluating Long-Term Memory of LLM-Based Agents in BIM Information Retrieval (2026) · cites this paper as [10]
Existing LLM-based BIM retrieval systems and benchmarks remain largely stateless.
“existing LLM-based BIM retrieval systems and benchmarks remain largely stateless”
From IFCMemoryBench: Evaluating Long-Term Memory of LLM-Based Agents in BIM Information Retrieval (2026) · cites this paper as [10]
The citation lists haven’t been fetched yet
Cited by 5
- Authors
- Sylvain Hellin, S. H., Stavros Nousias, S. N., & André Borrmann, A. B.
- Venue
- Computing in Construction
- Published
- 2025-07-14
- Volume
- 6
- ISSN
- 2684-1150 (ISSN-L)
- OpenAlex
- W4413862959
- Cited by
- 5
- References
- 0
- Citations by year
- 2026: 5
- FWCI
- 16.62
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- Computing in construction (publishedVersion)
- Topics
- Physical Sciences › Engineering › Building and Construction › BIM and Construction Integration
- Physical Sciences › Computer Science › Artificial Intelligence › Topic Modeling
- Physical Sciences › Engineering › Building and Construction › 3D Modeling in Geospatial Applications
- Keywords
- building information modeling, natural language interfaces, information retrieval, large language models, multi-agent systems, IFC standard, structural queries
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
