AI-Driven Identification of Reference Projects for Architectural Tenders: A Data-Driven Approach : Development of a Project Retrieval System and its Application in the AEC Industry

Sammanfattning

The identification of suitable reference projects is a critical yet time-consuming aspect of the architectural tendering process. This thesis investigates how arti- ficial intelligence (AI) can be leveraged to automate and optimize this task, fo- cusing on Cedervall Arkitekter as a case study. A data-driven retrieval system was developed to mine internal datasets—specifically the Milltime database— encompassing both structured project metadata and unstructured user notes. After evaluating multiple AI methods, an embedding-based retrieval approach integrated with keyword filtering was selected, striking a balance between computational efficiency and retrieval accuracy. Deployed on-premise as a web application, the final solution enables ar- chitects and procurement staff to query project records using natural language inputs. The system applies semantic similarity modeling and a customized ranking algorithm to provide rapid, relevant search results, cutting manual search time by more than 50% according to user testing. Structured interviews further demonstrated its capacity to enhance the reference project selection process and reduce reliance on personal memory. Taken together, these find- ings underscore the value of AI-driven retrieval systems in architectural prac- tices, while highlighting promising directions for expanded machine learning integration within tendering and other knowledge-intensive workflows in the architecture, engineering, and construction (AEC) sector.