28th International Congress on project Management "Project Management in the Digital Decade"

Project managers globally face ongoing difficulties in effectively managing and keeping pace with the vast amount of data generated on a daily basis throughout the lifecycle of construction projects. Large infrastructure projects generate thousands of letters, minutes of meeting and change orders, whose contractual value hinges on subtle layout cues – logos, stamps, marginalia. Recent studies show that up to 37 % of positional information is lost when construction PDFs are parsed with mainstream extractors, leading to expensive re-reads in claims and delay analyses. At the same time, large-language models (LLMs) are maturing as practical tools for contract analysis and project-controls tasks. This paper argues that pre- structured prompts – simple, human-authored schemas embedded in the user query – enable multimodal LLMs to turn chaotic correspondence into query-ready data.