Thomas Fel

Thomas Fel is a Member of Technical Staff at Goodfire and a Research Affiliate at Harvard University, where he previously served as a Kempner Research Fellow. He holds a Ph.D. in interpretable AI from Brown University and Université Paul Sabatier Toulouse III, and his research focuses on the interpretability of large vision models, particularly the geometric structure of neural representations. He is known for developing explainability tools including the Xplique library and the Sobol attribution method, as well as work on concept-based explainability frameworks such as CRAFT.