Data science & ML · Python track · hidden-holdout graded

Learn ML like an engineer.Scripts, not notebooks. Hidden holdouts. No videos.

Two courses that take a working developer from pandas and statistics to production model services, LLM/RAG evaluation and MLOps. Every exercise is graded by real test suites on hidden holdout sets — your model has to actually generalize, not just fit the cells you can see. Built for people who already program.

ds-ml-1 · evaluation · holdoutpytest · scikit-learn
# Train a churn classifier and report its F1.
# The grader re-runs this against a hidden holdout set —
# a model that memorized the training data fails the test.
def train_and_evaluate(train_df, test_df):
    X_train, y_train = split_features(train_df)
    X_test, y_test = split_features(test_df)

    model = Pipeline([
        ("scale", StandardScaler()),
        ("clf", LogisticRegression(max_iter=1000)),
    ])
    model.fit(X_train, y_train)

    preds = model.predict(X_test)
    return model, f1_score(y_test, preds)
--- PASS: test_train_and_evaluate · holdout F1 above threshold · ok
DS 1

Foundations

Data science for people who already program — wrangle real data, reason statistically, train scikit-learn models and evaluate them honestly. Scripts and hidden holdouts, not notebooks.

DS 2

Production, LLMs & MLOps

Ship models as tested services, build and honestly evaluate LLM/RAG features, and run MLOps-lite — tracking, versioning, drift monitoring. CPU-only, hidden-holdout graded.

×8

Capstones that ship

Eight capstones across the track — a data-cleaning gauntlet, an A/B test analyzer, a churn predictor, a model-as-a-service, a docs Q&A with a report card, and a drift hunt called Silent Failure.

01

The holdout decides, not vibes

Your model is scored against a hidden holdout set by real test suites. Pass/fail is deterministic — overfitting fails, and the AI never grades you.

02

It adapts to you

Struggle on a concept and the game diagnoses why and builds targeted practice. Already know pandas? It moves you up instead of making you sit through it.

03

Ship real ML systems

Capstones like a churn predictor and a document-intelligence service, with a senior-engineer code review — the answer to notebook copy-paste courses. Ship a portfolio, not just streaks.

Your first passing holdout is minutes away.

Start the DS & ML track — free

Free tier is genuinely free — every banked exercise is playable. We gate AI help and fresh generation, never the content.