Topics 01–03: ML pipeline, learning paradigms, OLS assumptions, regularization, sigmoid decision boundary, and data leakage traps. Tests whether you understand the why behind foundational models — not just their names.

Topics 04–06: Gini vs entropy, pruning vs overfitting, bagging mechanics, OOB error, residual fitting, learning rate, and the real differences between XGBoost and LightGBM. High trap density — essential for any ML interview.

Topics 07–09: Margin maximization, kernel trick, curse of dimensionality, k-selection traps, Laplace smoothing, and when the independence assumption breaks — or surprisingly doesn't. Tests distance-intuition and probabilistic reasoning.

Topics 10–11: Eigenvectors, variance explained, PCA leakage traps, when t-SNE lies, K-means convergence guarantees, DBSCAN parameter sensitivity, and silhouette score pitfalls. Essential for any role involving unsupervised pipelines.

Topics 12–13: Isolation Forest mechanics, LOF vs OCSVM tradeoffs, autoencoder anomaly scoring, bagging vs boosting vs stacking distinctions, and when ensembles actively hurt. Tricky evaluation scenarios included.

Topics 14–16: AUC-ROC vs PR curve, F1 tradeoffs, train vs test gap diagnosis, learning curves, L1 sparsity mechanics, Ridge vs Lasso selection, and dropout as regularization. Tests the reasoning behind metric and regularizer choice.

Topic 17: Filter vs wrapper vs embedded methods, mutual information limits, SHAP-based selection, categorical encoding traps, missing value imputation bias, and feature leakage detection. The topic most commonly underestimated in ML interviews.

Simulates an entry-level ML screening round. Broad coverage of 10 topics — fundamentals, supervised models, and evaluation. Every question has a trap designed to catch surface-level memorization. Pass this to prove you understand ML, not just its terminology.

Simulates a second easy ML screening — different question set, different topics. Covers unsupervised models, probabilistic classifiers, and regularization basics. Ideal as a second attempt or a complementary drill to Round 1.

Simulates a mid-level applied ML interview. Each question tests multi-step reasoning — design choices, debugging production models, and tradeoff analysis. Applied across 12 topics including boosting internals, kernel choices, and metric selection under class imbalance.

Second applied ML interview simulation — entirely different question set from Round 1. Covers NB vs LR tradeoffs, boosting depth choices, clustering evaluation, ensemble diversity, and feature selection under collinearity. Tests whether you reason from first principles.

Simulates a senior ML engineer interview round. 15 hard questions across all major algorithm groups. Expect edge cases, non-intuitive behavior, and scenario traps that break if you're reasoning from memorized patterns rather than model internals. FAANG-calibre difficulty.

Second hard mock interview — completely fresh question set. Tests architectural reasoning, optimization tradeoffs, debugging thought process, and production failure modes. Covers regularization effects, kernel behavior, anomaly evaluation under imbalance, and ensemble collapse conditions.

18 questions spanning all 17 ML topics at maximum depth. Designed for senior ML engineers, AI architects, and staff-level screening. Requires multi-step reasoning, internals knowledge, and production intuition. Questions test system tradeoffs, optimization failure modes, and edge cases that expose gaps in mental models — not surface-level recall.

Second elite assessment — entirely fresh question set across all 17 topics. Focuses on a different axis of depth: where Set 1 tests internals, Set 2 stresses architectural decisions, cross-topic interactions, and failure reasoning. Targeted at AI architects, ML platform engineers, and senior candidates who want to benchmark themselves against the hardest possible questions.