Fatima Zahra Hannou

Skills: Visualisation de données · Google Cloud Platform (GCP) · SQL

Main missions : • Researched and developed a post-processing module to correct OCR errors in Arabic documents using large language models (LLMs). • Simulated common OCR errors to create a realistic training dataset. • Fine-tuned Arabic LLMs such as AceGPT and Jais for OCR post-processing, achieving significant improvements in BLEU, WER, and CER scores. • Applied quantization techniques (GPTQ, AWQ, LoRA, QLoRA) to optimize model performance. • Proposed future improvements including few-shot learning and integration of Visual Question Answering (VQA) systems for smarter document understanding. • Leveraged SimLab HPC infrastructure for large-scale model training and evaluation under compute constraints.

Main missions : • Designed and implemented an industrial part recognition system based on one-shot learning and Siamese Neural Networks to classify components from a single reference image. • Built a full-stack web application using Vue.js (frontend) and Django (backend) enabling real-time image upload, training, and inference. • Integrated the solution with industrial cameras (GigE Basler) for automated image acquisition and real-time prediction in a production-like environment.

Main missions : • Began the internship by exploring and observing all sections of the washing plant (laverie) to understand the overall phosphate processing workflow. • Monitored and analyzed the performance of three industrial flotation units to assess phosphate enrichment efficiency. • Collected and tested samples at key processing stages (attrition, flotation entry, flotation exit) to evaluate BPL gains and MgO reductions. • Conducted root cause analysis (5M, 5 Whys, Ishikawa, SIPOC) to identify factors contributing to reduced flotation efficiency. • Acquired hands-on understanding of key industrial processes including attrition, conditioning, chemical treatment, and slurry transport.