Kush R. Varshney
IBM Fellow directing trustworthy AI research at IBM; team behind AI Fairness 360 and Granite Guardian
- IBM Fellow, directing Human-Centered Trustworthy AI research, IBM Research21
Kush R. Varshney is an IBM Fellow at IBM Research who directs Human-Centered Trustworthy AI research, and was the founding co-director of the IBM Science for Social Good initiative from 2015 to 2023. Varshney's homepage and IBM Research profile say Varshney and team created the open-source toolkits AI Fairness 360, AI Explainability 360, Uncertainty Quantification 360, and AI FactSheets 360, as well as the Granite Guardian model family. Varshney is a co-author of the papers introducing AI Fairness 360, AI Explainability 360, FactSheets, and Granite Guardian, and published the book "Trustworthy Machine Learning" in 2022.126893
Work
Granite GuardianOpen weights
Co-author of the Granite Guardian technical report (2024), which introduces risk-detection models for LLM prompts and responses released as open source; the ibm-granite/granite-guardian repository is under the Apache-2.0 license, and the Granite Guardian 3.0 8B model card lists the apache-2.0 license. Varshney's homepage lists the Granite Guardian model family among the work of Varshney's team.3451
AI Fairness 360 (external site: github.com)Open-source code
Co-author of the 2018 paper introducing AI Fairness 360, an open-source Python toolkit for algorithmic fairness released under the Apache v2.0 license; the homepage lists it among the toolkits Varshney's team created.671
AI Explainability 360 and FactSheets (external site: arxiv.org)Published research
Co-author of the 2019 paper introducing the AI Explainability 360 toolkit and of the 2018 FactSheets paper on documenting AI services.89
Other documented roles
Roles the sources document but that are not confirmed as current.
- Founding co-director (2015–2023), IBM Science for Social Good21
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.