At ITWORX Education, I built a four-stage LLM pipeline on the Gemini API that turns PDFs, webpages, and YouTube videos into structured competency lists. In one run, it reduced a 194-page book to eight categories and 14 competencies.
I added a verification step for model output and a tool that compares job requirements with a person’s skills to return a match score and learning plan. I also improved quiz generation in a second chatbot by organizing its output into a knowledge context layer.
For my graduation project, I classified sesame leaf diseases from photos using a ResNet50 backbone and a Kolmogorov-Arnold Network head. The model reached 98% mean accuracy across five-fold cross-validation, and I co-authored a paper that is being prepared for submission.
I also merged five patches into the LibreOffice codebase, modernizing legacy C++ and removing outdated macros. My other project compared YOLO models for blurring Egyptian vehicle license plates, with YOLOv11n achieving 0.9955 mAP@50.
