Abir Harrasse is a PhD student at ETH Zurich (D-INFK), advised by Prof. Mrinmaya Sachan and Prof. Bjoern Menze, and supported by the ETH AI Center Doctoral Fellowship. Her research focuses on understanding how deep neural networks work, when they generalize, and how to make them more reliable and interpretable, with specific interests in AI reasoning, robustness, and interpretability. Previously, she was a Research Scientist at Martian Learning Inc., focused on mechanistic interpretability and LLM routing. She completed a Diplôme d'Ingénieur from EMINES (first-class honors, best thesis award) with a minor in data science, and interned at the Max Planck Institute for Intelligent Systems under Zhijing Jin and Bernhard Schölkopf on multilingual LLM interpretability and training data attribution, and at the National University of Singapore on LLM evaluation frameworks.