At INSA Lyon, I’m studying in the M1 Master MINDS in AI program and working on classification, missing-data imputation, and LLM fine-tuning. I’m currently testing parameter-efficient methods to adapt pretrained models to specialized tasks.
For my toxic speech detection project using the HerWILL Dataset, I trained an SVM classifier and separately tested knowledge distillation from a transformer model, reaching approximately 50% accuracy. I traced the accuracy ceiling to label noise and class imbalance and established dataset-debugging protocols before hyperparameter tuning.
My background also includes network and operating systems studies at HIAST, where I ranked 8th in my cohort. I’ve analyzed Linux system logs and authored incident-response playbooks, and I currently teach mathematics remotely as a self-employed instructor.
