AI / ML
Lab Curriculum.

BUG FOUND: ML learners who can train a model on clean data, but cannot debug one in production.

$ step-01 brushup // weeks 1–2 — Topics taught:

1. Python, NumPy/pandas, and train/test splitting done correctly.
2. Supervised learning basics: classification, regression, and baselines.
3. Evaluation metrics: accuracy, precision, recall, F1, ROC-AUC, and when each misleads.
4. Data preparation: missing values, encoding, scaling, and leakage risks.
5. Overfitting vs underfitting, and simple validation (cross-validation).
6. Model serving basics: inputs, outputs, versioning, and logging predictions.

$ step-02 debug // weeks 3–4 — Production Debugging. Tools: Python, scikit-learn, Jupyter, evaluation scripts, prediction logs.

• BUG #1: Model accuracy collapses on real inputs — find data leakage in training, retrain cleanly, show honest metrics.
• BUG #2: A model returns nonsense for one segment — find the data / feature bug for that segment, fix it, show segment metrics.
• BUG #3: Drift: performance drops over time — detect drift in inputs, diagnose the cause, propose a retraining trigger.
• BUG #4: An AI feature hallucinates / fails silently — add validation, fallback, and logging so failures are visible, not silent.

$ step-03 navigate // weeks 5–7 — Codebase / System Navigation:

• Trace a prediction to data, features, model version, and serving code in a training + serving codebase.
• Fix a feature bug without breaking serving: version the change, test old vs new behaviour, roll back cleanly.
• Audit a model for a hidden data problem (duplicates, label errors, sampled bias) and document the finding.
• Review a peer's evaluation: check for leakage, wrong metric choice, and test-set contamination.

$ step-04 ship // weeks 8–9 — Shipping Features: capstone — ship an evaluated AI feature with metrics, error analysis by segment, a monitoring note, and a demo on real-style inputs.

$ ship → PROOF — exact output: evaluated model / feature, a metrics report (overall + by segment), an error-analysis note, and a monitoring / retraining note.

$ run --lab ai-ml → ✓ SHIPPED, job-ready