An independent applied research project
Archive Discovery Lab is an independent applied research project exploring whether private, source-linked search assistants can help archives make digitized photos, documents, transcripts, metadata, and finding aids easier to discover.
We are currently speaking with archivists and cultural heritage professionals to understand where search still breaks down after collections have been digitized — and whether plain-language discovery tools can reduce repetitive reference work without replacing archival expertise.
01 · Motivation
Many archives have invested years into digitization, metadata, finding aids, and online search tools. But researchers may still struggle when they do not know the right keywords, collection names, historical spellings, date ranges, languages, subject headings, or metadata structures.
When that happens, staff often become the search layer.
Archive Discovery Lab is studying whether plain-language, source-linked search can help staff and researchers move from vague questions to relevant candidate materials faster, while preserving source context, access restrictions, and archivist control.
02 · Origin
Research note
The first version was built to answer one archive’s real reference questions — not to demonstrate a product.
The project began with a real archival search problem.
William Jeffries was working with the Piłsudski Institute while researching materials related to Polish citizenship and family history. The archive had a large digitized collection, but finding relevant people, places, documents, and images across the material was still difficult and time-consuming.
He built an internal search system that ingested the digitized archive, created searchable transcripts and descriptions, and allowed staff to ask plain-language questions across the collection. The result was a working prototype that could surface relevant documents, images, metadata, and source links much faster than manual search alone.
Archive Discovery Lab now exists to learn whether similar discovery problems appear across other archives — and what a responsible, useful version of this tool would need to look like for archivists.
03 · Inquiry
Before assuming what archives need, we are speaking with archivists about the workflows, constraints, and trust requirements around AI-assisted discovery.
Q·1
When do researchers still need staff help after material has been digitized?
Q·2
Which search failures come from missing metadata, and which come from researchers not knowing the right archival language?
Q·3
Can plain-language search help with names, places, events, date ranges, language variants, and loosely described topics?
Q·4
Can image discovery help staff find relevant photos for researchers, exhibitions, classrooms, blog posts, and public programming?
Q·5
What source links, uncertainty labels, and review controls are needed before archivists trust AI-assisted discovery?
Q·6
Which materials should remain internal-only, restricted, or excluded entirely?
04 · Current work
The current prototype creates a private search layer over selected digitized archival materials. Staff can ask plain-language questions and receive candidate records, images, snippets, metadata, transcripts, and links back to the original source materials.
The system can work with digitized text, scanned documents, PDFs, images, handwritten documents where legibility allows, metadata, and transcripts from audio or video. It can support multilingual search and translation, and can create searchable descriptions when existing OCR or metadata is limited.
The prototype is designed as a discovery aid, not an authoritative replacement for archival description.
Workflow
A researcher asks for material about a person, place, event, community, organization, or time period. The system retrieves candidate records and source links that staff can review before responding.
Workflow
Staff search for images using descriptive, visual, or contextual language — for example, materials useful for exhibits, classrooms, social posts, rights and reproductions, or researcher requests.
Workflow
The system can help surface related materials when names, places, organizations, or topics appear in different languages, spellings, or historical forms.
Workflow
Staff ask broad questions across a collection to understand recurring people, places, topics, dates, and themes.
Workflow
The system can create searchable text from scans and documents when existing OCR is incomplete or unavailable. Existing OCR can also be used when preferred.
Workflow
With staff review, the system can help draft research summaries, collection notes, blog-post ideas, captions, and other public-facing material grounded in source records.
A short prototype example shows how a staff member might ask a plain-language question and receive source-linked candidate materials for review.
View prototype example05 · Boundaries
Archive Discovery Lab is not trying to replace archivists, catalogers, finding aids, or collection management systems. The project is focused on discovery workflows that remain grounded in source materials and human review.
06 · Participation
We are looking for a small number of archives with digitized collections to participate in a no-cost research pilot.
The goal is to learn from real archival workflows: what works, what fails, what staff would trust, and where AI-assisted discovery may or may not reduce repetitive search and reference work.
07 · Principles
Archives steward materials that carry context, rights, restrictions, community meaning, and historical uncertainty. Any AI-assisted discovery system must be designed around those realities.
Archive Discovery Lab is especially interested in how archivists think about source context, restricted materials, sensitive collections, metadata bias, copyright, cultural protocols, and the limits of machine-generated interpretation.
Results should point back to original records, images, transcripts, metadata, or collection pages whenever possible.
The prototype is designed to support staff review, not bypass it.
Partner archives decide which materials are included, excluded, internal-only, or potentially public-facing.
For each pilot, data access, model/API use, retention, exclusions, and deletion expectations should be agreed in advance. Public collections, partner-provided access, and local/private deployment options can be discussed depending on the archive’s needs.
The current implementation can use third-party AI model APIs for processing and search. For collections that require stricter controls, local or institution-controlled deployment options can be discussed.
The system should communicate uncertainty and avoid presenting generated text as archival authority.
08 · Context
Archive Discovery Lab is independent, but the project sits within a broader conversation about responsible AI, machine learning, and computational methods in libraries, archives, museums, and digital humanities.
These links are provided as field context only. Archive Discovery Lab is independent and is not affiliated with these organizations unless otherwise stated.
09 · People
Archive Discovery Lab is led by Evan Kimbrell and William Jeffries.
The project is currently in an applied research phase. We are speaking with archivists and cultural heritage professionals before deciding what the system should become.
William Jeffries is an AI engineer who began building the first version of the system while working with the Piłsudski Institute and searching through digitized historical materials related to Polish citizenship and family history.
Evan Kimbrell is an online educator and builder with experience turning complex subjects into useful learning systems.
Both are history enthusiasts interested in how new tools can support access to archival materials without weakening the role of archivists, source context, or institutional stewardship.
10 · Correspondence
If you work with a digitized archive and are willing to discuss how researchers currently find material, we would be grateful to hear from you.
We are especially interested in archives with digitized photos, documents, oral histories, newspapers, metadata, finding aids, or mixed-media collections where discovery still requires staff mediation.
Email the project at
info@archivediscoverylab.org