NVIDIA·AI Engineer·Technical Phone Screen
May 2026
Concurrency-heavy technical screen for an AI Engineer role at NVIDIA. The whole conversation was basically an open-ended deep dive into concurrency primitives and systems pitfalls, no leetcode, just talking through concepts and tradeoffs for a while.
- What is the difference between concurrency and parallelism, and why does the distinction matter in practice?
- Walk me through the tradeoffs between threading, multiprocessing, and async I/O. When would you choose each?
- How do race conditions occur, and how have you dealt with them in a real system?
- Explain the difference between a lock, a mutex, and a semaphore. When does each one apply?
- What is a deadlock, and how does it differ from a livelock? How do you detect or prevent each?
- What are lock-free data structures and what problems do they solve? What are the downsides?
- How do memory models and atomicity affect concurrent program correctness? What guarantees does a language or hardware memory model actually provide?
- What are some common concurrency pitfalls you've seen or would watch for in production AI systems?
“Felt good about this one.”