Applied Digital Curation: Engineering Multi-Agent AI Pipelines for Structural Data Normalization

Executive Summary: This advanced field report details the expansion of my role into digital asset management, showcasing the design and orchestration of an 8-page, multi-agent AI metadata pipeline. By programming and stress-testing three distinct AI persona agents (Architectural Historian, Senior Archivist, and Museum Docent), this project successfully scaled up cataloging consistency and enforced rigid input validation rules across thousands of historical archive records.

Antonio Torres-Moreno

June 14, 2026

Chula Vista Public Library Civic Center Branch

120-Hour Internship Progress Reflection

At the 120-hour mark, my role as a Digital Archives Intern has evolved to encompass the responsibilities of a Digital Asset Specialist, Metadata Technician, and Historical Reference Assistant. I am simultaneously operating at the intersection of information science, local history, and AI prompt engineering. Although my primary task continues to be digitizing the John Rojas Photo Collection, the skills I have developed through my own initiative have made me a far more effective archivist for the Chula Vista Public Library Civic Center archives.

My day-to-day responsibilities have shifted from simple data entry to engineering an automated, multi-agent metadata pipeline. This process was especially challenging because of the constant rigor involved in optimizing and testing the AI constraints for my specific tasks. My major accomplishment during this block was successfully expanding my original three-page system, which relied on a single ‘Senior Archivist’ persona to clean data, into a unified, nearly eight-page multi-agent system. By orchestrating three distinct specialized AI personas (an Architectural Historian for style classification, a Senior Archivist for structural data normalization, and a Museum Docent for public-facing narratives), I successfully scaled up our cataloging consistency. This iterative optimization process has taught me highly transferable technical skills in AI systems orchestration, including algorithmic thinking, data constraint engineering, input validation, and adversarial stress testing (Red Teaming) in creating a successful master prompt.

In tandem, I have also been developing my skills in archival reference services. My supervisor, the local history librarian, now forwards photo requests from purchasing clients directly to me. I am tasked with locating the relevant physical photographs across the collection's dozens of boxes, digitizing them using a specialized CZUR scanner if they aren't already online, and compiling the descriptive information into an inventory for the clients. I am currently fulfilling my second reference request, where the historical images I locate are being selected for potential use as decor in an upcoming breakfast restaurant chain nearby in Chula Vista CA.

Finally, this work has dramatically accelerated my ability to recognize and identify Southern California vernacular architecture. I have learned to navigate the nuances of architectural hybridity, accepting that few historic structures reflect a single, "pure" style, and instead to rely on isolating visual clues from specific architectural identifiers. Looking forward to the next phase of my internship, my goal is to further streamline our client reference workflow, ensuring that as public photo requests increase, I can rapidly query our newly standardized database to retrieve assets efficiently without disrupting the core digitization pipeline. While programming the AI tools has been an invaluable asset for navigating ambiguous edge cases, I have learned that the system ultimately relies on the precision of my own independent historical research and visual literacy.