Hands-on ML, deep learning and NLP from the Hebrew University AI Engineers course.
Production AI work rests on fundamentals. I'm building that foundation formally, in a 210-hour course, alongside shipping products.
Tabular, text and face-image models in PyTorch. The tabular MLP reached 0.85 ROC-AUC against a 0.78 XGBoost baseline. Compared a CNN built from scratch with pretrained FaceNet, ran leave-one-modality-out analysis with bootstrap confidence intervals, and debugged a CNN that wouldn't learn (missing BatchNorm).
EDA and preprocessing on telecom data; logistic regression, random forest, XGBoost and CatBoost baselines; overfitting analysis and tuning; model explanations with SHAP.
Data in SQLite via SQLAlchemy, exploratory analysis in pandas, and linear regression with OLS diagnostics.
Text cleaning, TF-IDF document similarity, BM25 retrieval and LDA topic modeling on a text corpus.
Course in progress (210 hours): classic ML, deep learning, deploying models, autonomous agents and NLP.