Cinema’s Agricultural Entanglements: Using LLMs to Map the Global Photochemical Supply Chain

The histories of iconic film manufacturers such as Kodak, Agfa, Gevaert, Fujifilm, and Konica remain poorly connected to the global supply chains that made photochemical cinema possible. The global networks, convoluted corporate histories, and the labor and extraction are difficult to reconstruct because collection are fragmented, unevenly preserved, rarely digitized, and intensely multilingual.

Combining traditional archival research with computational techniques, the CINEAGRI project has assembled a multilingual corpus of more than 30,000 pages of records related to the history of photochemistry. The goal is to build a “map of celluloid” that emplaces raw materials within their regional, political, and environmental contexts.
This paper presents a number of provisional solutions developed by CINEAGRI in response to the methodological challenge of operationalizing this unique collection through LLM-assisted OCR, document clustering, metadata extraction, retrieval-augmented generation, and knowledge graph construction. Japanese sources – including handwritten drafts of corporate histories, and tabular data in “exotic” layouts – serve as a stress test for current models and workflows. The performance of several commercial models in various tasks related to historical research (especially structured data extraction) is compared.

I argue that RAG and knowledge graphs can effectively support historical supply-chain research (as well as knowledge sharing in alternative formats) when designed around provenance and uncertainty. This workflow reveals obscure suppliers, material substitutions, personnel movements, and “chemical borders” when film materials change identity.