531,459 interview questions from 6,000+ companies.
Tests prioritization under pressure, stakeholder management, and ownership when multiple urgent requests compete for limited time.
Tests influence without authority through stakeholder alignment, communication, and ownership in a high-stakes decision.
Tests how you handle ambiguity while maintaining accuracy, documentation discipline, and ownership of the final output.
Design an LLM serving system that balances latency, cost, scalability, and safety for production traffic.
Explain how to evaluate a generative model using offline and online methods, with attention to hallucination, product metrics, and experiment design.
Compare RAG and fine-tuning, and decide when each is the better fit for an LLM product.
Design a low latency RAG system over millions of documents, with scalable retrieval, ranking, generation, and production monitoring.
Design a production agent platform that coordinates models, tools, and data sources under strict latency, cost, and safety limits.
Tests your ability to evaluate embedding models and retrieval quality in production settings.
Tests your ability to build reliable RAG evaluation metrics and experimental methodology.
Tests your understanding of normalization layers and your ability to implement them correctly.
Tests your grasp of positional encoding math and how it affects transformer behavior.
Tests your ability to improve performance using async patterns and concurrency control in Python.
Tests your knowledge of distributed training memory optimizations and their practical implications.
Tests your ability to implement core transformer attention correctly and efficiently in PyTorch.
Tests your understanding of distributed training strategies and when to use each.
Tests your ability to implement MoE routing and manage expert utilization effectively.
Tests your debugging skills and understanding of transformer normalization order.
Tests your understanding of distributed systems performance techniques for training.
Tests your understanding of attention kernel optimizations and memory bandwidth bottlenecks.
32 total questions
