Case Studies

Newsagents: AI research with sources attached

Newsagents helps students, scholarly researchers and community historians explore primary sources, including WWII-era newspapers, books and journal articles. Every AI answer links back to the original scan.

Julien Gorbach, a journalism professor at the University of Hawaii, asked Montana Banana to build the backend pipeline and an interface that would work for first-year students as well as experienced researchers. We built the platform around his carefully selected archive, with OctoberCMS managing ingestion, agents and content.

Challenges

Newsagents homepage offers guided topic browsing and a direct route into the archive.

A useful research tool has to do more than produce a convincing answer. Researchers need to find the evidence and check it for themselves.

Uneven scans, one searchable archive
The collection included microfilm scans of varying quality, books and journal articles. Each source needed to become structured, searchable material without losing its connection to the original.
Evidence a scholar can verify
An answer that sounds right isn’t enough for scholarly work. Researchers needed to retrieve and cite real sources, then open the original scan to check each claim.
A starting point for new researchers
The interface needed to support questions across different disciplines while remaining approachable enough for a first-year student to use before a class assignment.

Approaches

Newsagents Browse Topics interface offers People, Issues, Events and The Black Press research paths.

We built the research pipeline around traceable sources, then gave users several ways into the archive.

Structure the sources first
OCR and parsing turn scanned pages into structured, tagged article records before the language model sees them. OctoberCMS keeps ingestion, agents and content in one backend.
Retrieve before generating
A hybrid Typesense pipeline combines keyword search with semantic reranking. The model receives a capped set of relevant documents and instructions to answer only from those sources. Responses link to the original scans, and CMS-managed synonyms help connect different phrasings to the archive.
Make the archive approachable
A mobile-first, 1930s-newsroom interface offers a Browse Topics wizard and named agents with different disciplinary perspectives. Answers separate Analysis from Sources. A standalone search page lets researchers skip the AI entirely.

Results

Used in class

Dr. Gorbach assigns Newsagents to his students. They can explore the archive through questions and follow source links back to the scans behind each answer.

Conference demos

Dr. Gorbach demonstrates Newsagents at conferences and discusses the platform with publishers and archive partners. The working tool gives those conversations a concrete research experience.

Client-led growth

Julien expands the archive himself through OctoberCMS. Montana Banana continues to meet with him as the platform moves into its next phase of development.

Archives, meet AI.

Let’s turn it into a tool people can investigate.

Start Your Project