Material Classification - Doug Peterson Imaging Product Manager, Image Science, Product Evangelist, AI guru, and Photographer

Executive Overview

This project originated through the PA ArCHER Grant program that I conceived. This Grant generated 23 new leads (a big win in our high-LTV vertical) while providing a low-stakes way to explore AI's business value.

We awarded the grant to the Smithsonian. The resulting project demonstrates how AI can address critical metadata challenges for cultural heritage institutions, delivering immediate value while establishing reusable methodologies for future engagements.

As Head of R&D and Product Manager at Digital Transitions, I led both the Grant and the resulting AI project, with the following key results:

Improved Collection Access: Enhanced access to 112,000 digitized items that were previously "dark" due to lack of metadata
Workflow Efficiency: Reduced processing time from years to weeks, saving over 3,600 staff hours
Cost Reduction: Decreased post-digitization costs by more than $130,000
Business Compliance: Ensured 99.5% accuracy in identifying sensitive contract documents containing royalty agreements, on par with standard human review
Scalable Methodology: Established a repeatable process for applying AI to similar collections

Lorem ipsum

Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit in voluptate velit esse cillum.

Lorem ipsum

Our Approach

Collection Analysis: Assessed needs and characteristics of the Smithsonian’s Asch Collection
AI Strategy: Identified material type categorization as optimal application of AI
Custom Model Development: Trained specialized algorithms for seven material types
Risk Management: Created enhanced protocols for sensitive content detection

Client Challenge

The Smithsonian's Ralph Rinzler Folklife Archives faced critical issues with their Moses and Frances Asch Collection:

112,000 digitized images with two-thirds lacking descriptive metadata ("dark" content)
Manual processing estimated at 3,600 hours costing $131,863, plus oppourtunity cost
Need to keep private sensitive business documents containing royalty agreements

Solution Development

Assessment: Determined AI could automate material type categorization
Data Preparation: Selected seven priority categories and refined training data
Model Development: Trained ResNet on our refined training data
Implementation: Applied asymmetric error handling to prioritize sensitive content protection

Lorem ipsum

Key Results

95% overall classification accuracy across seven material types
99.5% accuracy identifying contracts containing sensitive information
Similar accuracy to human processing
$130,000+ cost reduction compared to manual processing
Months to weeks reduction in processing timeline

Value Delivered

Material Classification: Enabled filtering by document type across the collection
Content Protection: Secured sensitive business documents with specialized detection
Operational Efficiency: Saved 3,600+ staff hours and redirected talent to higher-value tasks
Methodology Transfer: Established framework for applying AI to additional collections

Lorem ipsum

Client Outcome

The project transformed the Asch Collection from partially accessible to fully searchable, with Smithsonian Folkways gaining efficient access to business-critical documents while protecting sensitive information.

Success Factors

Targeted a specific, high-value challenge with measurable ROI
Tailored implementation to unique cultural heritage requirements
Balanced accuracy needs with practical considerations for different content types
Created foundation for expanded AI applications beyond initial use case