Construction lags behind other industries in productivity growth and business digitalization, which is also reflected in the verification of consistency between delivery notes and invoices on construction sites. This process is still carried out mostly manually, which is time-consuming and error-prone, since different suppliers use inconsistent document formats. The aim of the thesis was to investigate whether this process can be automated using large language models and workflow automation tools, without requiring a separate template for each supplier. To this end, a model of an automated system was designed and a working prototype was built, which reads documents from a cloud folder, extracts data from them using a large language model, and automatically compares them in a spreadsheet. The prototype was tested on real delivery notes and invoices. It was shown that this approach reliably detects quantity discrepancies for goods with stable product codes, while challenges remain with less transparent line items, such as services, and with the risk that the model may fabricate a data point. For this reason, the system cannot operate fully autonomously and requires a human to occasionally review flagged cases. Beyond the technical solution, the thesis also addresses its limitations, including personal data protection concerns, the costs of processing at larger scale, and the placement of the solution within the broader trend of digitalization in construction. The findings show that such a system can substantially relieve the site manager of administrative work, even though it does not eliminate the need for occasional manual oversight.
|
