Generative Design Proposals via LLM-BIM Integration : master thesis

Khan, Sangeen (author), Turk, Žiga (mentor)

Large Language Models (LLMs) have unique advantages and potential in Building Information Modeling (BIM) environments. Their methodology supports natural language interaction with complex design software, offering efficient communication between AI systems and BIM-based platforms. However, for structural and architectural modeling workflows, the integration capacity of existing AIBIM approaches presents some limitations that must be resolved with best possible solutions. These challenges include unidirectional communication patterns and complex programming requirements that hinder implementing conversational AI workflows in BIM design offices. This study focuses to analyze the process of enabling bidirectional communication between LLMs and BIM software while developing a unique workflow for implementation within design tools, AI platforms, and project teams. Two case studies have been developed to test practical interoperability workflows: BIMCP for Autodesk Revit integration and Kolektor-BONCP for IFC manipulation through Blender's Bonsai addon. This is to identify and resolve the limitations involved in natural language command processing and data flow between AI assistants and the BIM software. The communication processes are bidirectional, using the Model Context Protocol (MCP) framework and socket-based connections for real-time interaction. AI-BIM integration and conversational design workflows are relatively underutilized. This is mainly due to interoperability problems and accessibility barriers. Despite the identified limitations and technical challenges, the study outcomes within the case study development illustrate how the workflow has excellent advantages in developing AI-assisted BIM projects and its implementation in construction industry design offices. Real-world validation at Kolektor Koling demonstrates measurable productivity improvements, transforming multi-hour manual processes into minute-long conversational queries.

• Language: English
• Keywords: master thesis, civil engineering, model context protocol, natural language prompts, building information modeling, AI-BIM integration, bidirectional communication, conversational interfaces
• Work type: Master's thesis/paper
• Typology: 2.09 - Master's Thesis
• Organization: FGG - Faculty of Civil and Geodetic Engineering
• Place of publishing: Ljubljana
• Publisher: [S. Khan]
• Year: 2025
• Number of pages: 1 spletni vir (1 datoteka PDF (XII, 57 str.))
• PID: 20.500.12556/RUL-171102
• UDC: 004.42:69(043.2)
• COBISS.SI-ID: 245263107
• Publication date in RUL: 05.08.2025

Secondary language

• Language: Slovenian
• Title: Predlogi generativnega oblikovanja preko integracije LLM-BIM : magistrsko delo

Abstract

Veliki jezikovni modeli (LLM) imajo edinstvene prednosti in potencial v okoljih informacijskega modeliranja gradenj (BIM). Njihova metodologija podpira interakcijo v naravnem jeziku s kompleksno programsko opremo za načrtovanje, kar omogoča učinkovito komunikacijo med sistemi umetne inteligence in platformami, ki temeljijo na BIM-u. Vendar pa za delovne tokove strukturnega in arhitekturnega modeliranja integracijska zmogljivost obstoječih pristopov AI-BIM predstavlja nekatere omejitve, ki jih je treba rešiti z najboljšimi možnimi rešitvami. Ti izzivi vključujejo enosmerne komunikacijske vzorce in kompleksne programske zahteve, ki ovirajo implementacijo pogovornih delovnih tokov umetne inteligence v BIM načrtovalskih pisarnah. Ta študija se osredotoča na analizo procesa omogočanja dvosmerne komunikacije med LLM-i in BIM programsko opremo, hkrati pa razvija edinstven delovni tok za implementacijo znotraj načrtovalskih orodij, platform umetne inteligence in projektnih ekip. Razviti sta bili dve študiji primerov za testiranje praktičnih interoperabilnostnih delovnih tokov: BIMCP za integracijo z Autodesk Revitom in KolektorBONCP za manipulacijo IFC preko Blenderjevega dodatka Bonsai. Namen je identificirati in rešiti omejitve, povezane s procesiranjem ukazov v naravnem jeziku in pretokom podatkov med asistenti umetne inteligence in BIM programsko opremo. Komunikacijski procesi so dvosmerni, z uporabo ogrodja Model Context Protocol (MCP) in povezav, ki temeljijo na vtičnicah, za interakcijo v realnem času. Integracija AI-BIM in pogovorni načrtovalski delovni tokovi so relativno premalo izkoriščeni. To je predvsem posledica problemov z interoperabilnostjo in ovir pri dostopnosti. Kljub identificiranim omejitvam in tehničnim izzivom rezultati študije znotraj razvoja študij primerov prikazujejo, kako ima delovni tok odlične prednosti pri razvoju projektov BIM s pomočjo umetne inteligence in njegovi implementaciji v načrtovalskih pisarnah gradbene industrije. Validacija v realnem svetu pri podjetju Kolektor Koling dokazuje merljive izboljšave produktivnosti, saj pretvarja večurne ročne procese v minutno dolge pogovorne poizvedbe.

• Keywords: magistrska dela, gradbeništvo, povpraševanja v naravnem jeziku, informacijsko modeliranje stavb, integracija AI-BIM, dvosmerna komunikacija, pogovorna vmesnika