Most ML courses teach concepts in isolation. GradCurve teaches them through a single real project — a building energy classifier — from raw CSV all the way to production deployment. Every lesson connects to actual source code, important formulas include relevant interactive visualizations, and a 58-question exam checks whether it stuck.
Students who finished an ML course and want to see how everything connects in one real codebase. Not a course replacement — a companion for when you are staring at someone else's code and wondering why it was written that way.
49 lessons across two tracks:
ML fundamentals — dataset loading, feature engineering, scaling, logistic regression, regularization, SVM, decision trees, kNN, naive Bayes, dimensionality reduction, clustering, neural networks, autoencoders, PyTorch, CNNs, RNNs, ensembles, gradient descent, overfitting
Production engineering — MLflow, FastAPI, Docker, Streamlit, GitHub Actions CI, AutoML, SHAP and LIME explainability, data validation, model cards, LLM streaming
14 interactive visualizations — move a slider and watch the formula value, the chart, and the model output all change at once.
5 browser demos — including a live in-browser classifier that trains three models on synthetic data with no Python server.
58-question final exam and a 101-term searchable glossary.
EnergyTypeNet: https://github.com/bartoszbryg/EnergyTypeNet
30 custom NumPy model implementations, 19 notebooks, FastAPI service, Docker deployment, MLflow tracking, Streamlit dashboard, SHAP and LIME explainability. Covers all 20 topics of a standard ML course syllabus plus the production engineering story.
pip install flask
python build.py
# Open http://localhost:5000No npm. No build step. Static HTML + CDN React 18 + plain JavaScript.
- Add a
NAVentry insrc/app.jsx - Add a
LESSON_IDXmapping insrc/app.jsx - Push content to
window.BLOCKS[n]insrc/lesson-content/*.js - Add the script tag in
index.htmlin load order
Built alongside the EnergyTypeNet ML portfolio project.