Retrieval across text, transcripts and metadata so a researcher finds the clip by describing it, with the source shown.
Automatic enrichment of incoming material, reviewed in bulk rather than item by item.
Interfaces built with the desk that uses them, for drafting support, versioning and fact-check trails.
Anything generated is bounded by what your licences permit and labelled so an editor knows what they are looking at.
Short cycles, visible progress, and a scope you can change. You see working software every week rather than a status report.
The repository, the infrastructure code, the evaluation set and the documentation. In your accounts, under your licence, from the first commit.
Time with the desk and the archive team, a sample corpus, and the questions people actually try to answer.
Working retrieval over a slice of the archive, measured against searches your researchers ran last month.
Enrichment pipelines, editorial interfaces and rights checks, rolled out to one desk first.
Corpus expanded, quality reviewed with editorial, and tooling extended as the workflow changes.
No automated publishing. Everything the system produces is a draft with a named human accountable for it.
Outputs carry their sources, because in this industry an unsourced sentence is a liability.
Tone, terminology and style guidance encoded from your own guide, not a generic one.
What may be trained on, generated from or republished is agreed in writing before we build.