Nour Karnib

This project demonstrates end-to-end applied AI engineering: starting with data preprocessing on messy Amazon reviews from Kaggle, I prepared noisy real-world text into a structured format. I then fine-tuned GPT-2 to adapt a foundation model to the task, while managing version control with large files — an often-overlooked but highly practical engineering skill. The model was deployed on Hugging Face Hub, making it accessible as an interactive service: https://huggingface.co/spaces/nourkarnib/fine-tuned-gpt2 Importantly, I approached this not just as model training, but as a problem-framing exercise, restructuring inputs to reduce perplexity and achieve measurable performance gains. https://github.com/nourkarnib/Gen-AI-Opinion-about-products-based-on-AWS-data

• Proposed a statistical method to identify key switches, reducing solver (CPLEX, Gurobi) computational time (up to 177x faster). This work improved decision-making efficiency in large-scale network optimization. • Focused on mathematical programming and optimization for large-scale Distribution Network Reconfiguration (DNR), a well-known NP-hard problem in operations research, with direct analogy to real-time delivery networks where supply and demand must be balanced efficiently. • Modeled critical constraints (voltage variations, current limits, etc.) in electrical networks, comparable to capacity and timeliness constraints in delivery systems, and validated solutions against exact simulation methods. • Designed innovative optimization solutions for networks with high renewable-energy integration, analogous to volatile and dynamic delivery demand, including approaches for curtailment and flexibility under peak conditions