United States of America · Washington, D.C. · See the United States of America profile · See the Washington, D.C. profile
Evidence: Descriptive / self-reported Top 12% 80/100 · Ask Evidence Copilot about this practice
Since around 2019, NASA's Scientific and Technical Information Program has used a machine-learning model — trained on 3.5 million tagged documents — to auto-tag NASA's open data against a taxonomy of roughly 20,000 standardized keywords, and open-sourced the tooling.
NASA's Scientific and Technical Information (STI) corpus spans scientific articles, technical reports, software repositories and other unstructured content scattered across multiple systems, historically tagged by hand with roughly ten keywords per document drawn from an evolving taxonomy of more than 20,000 standardised terms.
The STI team and NASA's OCIO Data Analytics Team set out to build a machine-learning and NLP-based automated tagging system to reduce the manual burden of tagging and improve discoverability of NASA's technical content.
The system was trained on the NASA Technical Reports Server's corpus of 3.5 million already-tagged documents, then released as open-source software (the concept-tagging-training and concept-tagging-api repositories on GitHub) and documented as a Federal Data Strategy 'proof point' case study, published May 2019. Its use has since expanded beyond original article-tagging to other unstructured NASA data.
No public accuracy, usage or before/after discoverability figures have been published, so the tool's measured impact on data findability remains undemonstrated even though the system has been operating for several years.
The case study explicitly frames the tool as a template other federal agencies can replicate on their own text corpora, backed by open-source code rather than a proprietary black box.
National / regional programmes
Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.
Reusable artefacts from this practice — as published by their sources.
Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.
Where this practice's information was retrieved from, and when.
United States of America
The US State Department's enterprise AI chatbot StateChat grew from 150 testers to 62,000+ users by June 2026; a 2025 …
United States of America
Facing a mandatory 25-year declassification deadline and a year of manual review per annual cohort, the U.S. State Department trained …
South Korea
Before agencies publish a dataset on Korea's Open Data Portal, a diagnostic service scores it on five criteria — understandability, …
Brazil
Brazil's civil-society Operação Serenata de Amor uses open-source machine learning ('Rosie') to flag suspicious deputy expense claims. It flagged ~8,000 …
Open full copilot Grounded in cited practices — always check the sources.