Most of my work on AI relies on understanding data — closing the gap between a notebook that works once
and a model that survives real-world data. But I always find myself questioning why it can't generalize at real-world scale,
why a model behaves like a bad human and cheats its way to a good train result.
And in the end, I concluded: the human mind will always stay a wonder, no matter what AI reaches.
AI Research Intern @ UM6P (Geology and Sustainable Mining Institute, Ben Guerir). Rebuilt a geosite-discovery pipeline after finding target leakage in the original, retraining it on 1662 expert-sourced labels with properly clustered cross-validation. Also modeling Cr(VI) adsorption on iron-grafted biochars.
AI/ML Engineering Intern @ FaceJob (remote). Built the AI pipeline for a video-CV recruitment platform from scratch, from Whisper transcription to LLM-based coaching, hardened against prompt injection.
| Publication | FSC-Net: Fast-Slow Consolidation Networks for Continual Learning (arXiv:2511.11707), a dual-network architecture that fights catastrophic forgetting on Split-CIFAR-10 |
| Competitions | Moroccan Collegiate Programming Contest (MCPC 2025), CODE IT V8 (EHTP), the NVIDIA Nemotron Reasoning Challenge on Kaggle, and Kaggriculture, a farming-sim capstone where a rule-based crop agent beat the baseline by roughly 4x |
| Certifications | Oracle Cloud Gen AI Professional, OCI AI Foundations Associate |





