Kartik Goyal#,1, Myrsini Mamoli#,1, Huopu Zhang1 , and Jingqi Ruan

We present a multi-stage pipeline for collecting, filtering, and structuring literature on Greco-Roman architecture, and in particular temples and libraries. The workflow spans (i) large-scale retrieval using author-name and keyword queries, (ii) initial relevancy scoring with a visual-language model (Colpali), (iii) supervised classifiers operating on titles and on sequences of detected figures/tables, and (iv) information extraction to produce building-level metadata (order, typology, chronology, location) with evidence snippets. We conclude with an interactive web map visualizing aggregated site/building data.