I am a final year (5th year) engineering student at EMINES School of Industrial Management, minoring in Data Science. My research interests lie in the design of next-generation AI algorithms that possess advanced reasoning capabilities. I am especially fascinated by the development of more efficient and intelligent AI systems, as well as interpretability methods that enhance our understanding of large language models (LLMs).

Currently, I am pursuing a research internship in mechanistic interpretability under the supervision of Zhijing Jin at the Max Planck Institute for Intelligent Systems, where I work on understanding and improving the internal reasoning mechanisms of LLMs.

Previously, I was introduced to Mechanistic Interpretability in my research intern role at WithMartian, where I contributed to two research papers exploring novel circuit analysis techniques for LLMs. Additionally, I explored novel evaluation frameworks for LLMs during my internship at NUS, drawing on social, economic, and voting theories to develop human-like, accurate, and cost-effective assessment methods.

In summer 2023, I interned at Africa Business School, supervised by Agnès Gorge, working on assortment optimization for major Moroccan retailers like Marjane.

I maintain a blog on medium where I share my insights on AI, and I actively participate in hackathons to craft innovative AI solutions. I won’t lie, I get a real thrill from winning them 😄.

Recent news:

• 🧪 April 2025: I’ve started a new research internship on mechanistic interpretability under the supervision of Zhijing Jin at the Max Planck Institute for Intelligent Systems in Tübingen! Excited to dive deeper into how LLMs work and how they can be aligned agaisnt inherent model bias.
• 📄 March 2025: Excited to share our new paper TinySQL: A Progressive Text-to-SQL Dataset for Mechanistic Interpretability Research is now on arXiv! We introduce TinySQL as a benchmark and apply techniques like edge attribution and sparse autoencoders to identify circuits behind SQL generation.
• 🔬 March 2025: Excited to see “Activation Space Interventions Can Be Transferred Between Large Language Models” out! I had a great time contributing to this project and learning from the team. It explores how safety interventions can transfer across models.
• 📄 October 2024: Thrilled to share that our paper “Adversarial Multi-Agent Evaluation of Large Language Models through Iterative Debates” is now live on arXiv! This work, co-authored with Chaithanya Bandi, explores a novel framework for evaluating LLMs through multi-agent debates.
• 🚀 September 2024: Excited to announce that I joined WithMartian, where I focused on LLM routing technology and mechanistic interpretability research. I had the opportunity to co-author two papers and contribute to building a powerful interpretability research toolkit.
• 🔍 April 2024: I joined NUS as a research intern, where I worked on novel evaluation frameworks for LLMs inspired by social choice and economic theory.
• 🎉 April 2024: We’re thrilled with the overwhelming success of the first edition of the Moroccan Day of Mathematics. It’s incredibly rewarding to see an event I co-founded with my friends at Math&Maroc make such a meaningful impact.
• 🌟 March 2024: The Winter School featuring prominent AI researchers like Yann LeCun and Eric Xing and that I proudly coorganized took place.
• 🏅 December 2023: Thrilled to share that our medical solution, SehhaTech, advanced to the finals of both the IoT & AI Challenge and the MoroccoAI hackathon.
• 🏆 October 2023: I had the pleasure of participating in the BCG Platinion Hackathon and winning first place. Our solution addresses a major aspect of mobility for grocery stores and also functions as a collaborative, optimized platform for emergencies. The idea was inspired by the need for better coordination in the recent earthquake relief efforts in Morocco.