Rəfi Veyisov

I’m a Machine Learning/AI engineer with a Master’s in Computer Science and a strong foundation in Python, TensorFlow, PyTorch, and MLOps best practices. I build, evaluate, and deploy end-to-end ML models through hands-on academic and personal projects.

In my projects, I’ve developed a web-based handwritten character recognition system using CNNs trained on 100,000+ EMNIST samples, delivering ~83% accuracy with real-time predictions and top-3 confidence visualization. I also built a movie review sentiment app comparing SimpleRNN vs LSTM, where LSTM reached 86–88% accuracy versus SimpleRNN’s 83–85% on the IMDB dataset.

I’ve focused heavily on classical ML fundamentals and production-minded model development: I tuned Decision Tree and Random Forest classifiers with Grid Search, Random Search, and Optuna, achieving ~98% ROC-AUC and finding best parameters ~2.5x faster than Grid Search. Alongside that, I compared NumPy-from-scratch vs scikit-learn regression (R² 0.87 vs 0.89) and ran rigorous MNIST evaluation with cross-validation, confusion matrices, precision-recall/ROC curves, and threshold tuning.