Question bank

4,310 interview questions, answered.

Reference answers, what the interviewer is really testing, how it is graded, and the follow-ups that come next.

Easy 7Medium 19Hard 36
Filters · onDifficulty, topic, company
Clear all

19 questions

  1. 1Context windows and their practical limitsMediumAI ML· Rr/ExperiencedDevs
  2. 2Why divide by sqrt(d_k) in scaled dot-product attention?MediumAI ML· OMOpenAI MLE
  3. 3Bias-variance tradeoff: practical rules of thumbMediumAI ML
  4. 4BPE vs SentencePiece vs WordPiece tokenizationMediumAI ML
  5. 5Dense vs sparse (BM25) vs hybrid retrievalMediumAI ML
  6. 6Diagnosing bias vs variance by experimentMediumAI ML
  7. 7Document chunking strategies for RAGMediumAI ML
  8. 8Grid search vs random search vs Bayesian optimizationMediumAI ML
  9. 9Handling severe class imbalance in productionMediumAI ML
  10. 10LoRA vs full fine-tuningMediumAI ML
  11. 11Model Context Protocol (MCP)MediumAI ML
  12. 12RAGAS: why it's popular and where it breaksMediumAI ML
  13. 13The ReAct frameworkMediumAI ML
  14. 14Top-k, top-p (nucleus) and temperature samplingMediumAI ML
  15. 15What is instruction tuning and why does it matter?MediumAI ML
  16. 16What is KV cache and how does it speed up inference?MediumAI ML
  17. 17Why do transformers need positional embeddings at all?MediumAI ML
  18. 18Why multi-head attention instead of one big head?MediumAI ML
  19. 19Word embeddings, from classic to modern retrievalMediumAI ML