314,552 interview questions from 6,000+ companies.
Describe how you handled a disagreement with an engineer or safety expert when the decision involved delivery pressure and safety tradeoffs.
Explain the bias-variance tradeoff and how it guides model choice, regularization, and generalization performance.
Reflect on a real execution failure, what caused it, how you responded, and what you changed afterward.
Design a grounded document Q&A system and explain how vector search improves retrieval quality, latency, and hallucination control in RAG.
Choose a decision threshold for a classifier using precision, recall, calibration, and confusion matrix tradeoffs.
Explain a practical process for tuning model hyperparameters using cross-validation and overfitting checks.
Explain how transformer self-attention works, including its role in sequence modeling and why it scales better than RNNs.
Choose sample size and runtime by combining baseline rate, MDE, alpha, power, and expected traffic.
Explain how to improve coding solutions by reducing time complexity first, then balancing space trade-offs.
Interpret what a 0.84 AUC-ROC means for a marketing response model and explain why threshold and calibration still matter.
Build a classifier for a rare-event problem and choose metrics and training tactics that work when positives are scarce.
Assess whether a ranking model's predicted click probabilities are trustworthy when ranking quality is stable but probability estimates are miscalibrated.
Explain how to evaluate whether predicted probabilities match observed outcomes, and how to interpret calibration in practice.
Assess whether a payment fraud model is calibrated well enough for auto-decline and review decisions despite strong AUC-ROC.
Build a transformer-based system that explains LLM mechanics for learners and classifies concept coverage across key topics like tokenization and attention.
Design and implement a grounded RAG pipeline that retrieves relevant context before generating answers.
Explain embeddings and how to apply vector search to semantic retrieval over medical text.
Adapt a language model to clinical notes using domain preprocessing, supervised fine tuning, and task specific evaluation.
Fine-tune a transformer to map clinical text to diagnostic categories with careful handling of imbalance and error analysis.
