OpenAI AI Engineer interview questions

4 AI Engineers have shared notes from their OpenAI interviews, with the questions they were asked in each round. That is 1% of the 667 OpenAI interview notes on Screna AI.

4 AI Engineer interview notes

OpenAI·AI Engineer·Technical Phone Screen

Jun 2026
OpenAI AI Engineer interview with a pretty gnarly math/probability question about optimal restart strategies for LLM inference. The whole thing was centered on one problem with layered sub-questions, and it got deep fast.
  • You have an LLM solving a math problem where the time to a correct answer follows a known heavy-tailed distribution. You can abort and restart at any time, with each attempt drawing independently from the same distribution. Can you design a restart strategy that achieves a lower expected total time than just letting a single attempt run to completion?
  • For a single-threshold restart strategy where you abort and restart whenever elapsed time exceeds a fixed threshold T, derive the expected total time as a function of T and find the value of T that minimizes it.
  • Design a strategy that strictly outperforms the best single-threshold policy in terms of worst-case performance. You can consider a deterministic schedule of multiple restart thresholds or a randomized restart policy.

“The setup took me a minute to internalize.”

View Post

OpenAI·AI Engineer·Technical Phone Screen

May 2026
OpenAI AI Engineer interview with a multi-part coding problem centered on matrix multiplication and automatic differentiation. The whole thing built progressively, each part depending on the last, which made it feel more like a gauntlet than a standard coding screen.
  • Implement an in-place matrix multiplication routine for a sequence of square matrices, then explain why this approach breaks backpropagation.
  • Rewrite the matrix multiplication chain as an out-of-place function that is compatible with autograd.
  • Manually implement the backward pass for the out-of-place matrix multiplication chain, computing gradients with respect to each input matrix without using autograd.
  • Implement both the forward and backward passes for the matrix chain product using a Hillis-Steele parallel prefix scan pattern, leveraging the associativity of matrix multiplication.

“The implementation itself was fine, maybe five minutes.”

View Post

OpenAI·AI Engineer·Technical Phone Screen

May 2026
Interviewed for an AI Engineer role at OpenAI and got hit with a pretty gnarly algorithmic scheduling problem. The whole thing revolved around generating balanced sequences of triples across multiple dimensions, which is not the kind of thing you can just wing with a standard LeetCode pattern.
  • You have three tag dimensions (AI, Human, Task) each with a finite set of values. Design an algorithm to generate a sequence of N triples, one from each dimension, such that at every prefix of the sequence, the count of any pair of values across dimensions stays as balanced as possible (max count minus min count is at most 1). Walk through your approach.

“This one took me a minute to even parse.”

View Post

OpenAI·AI Engineer·Technical Phone Screen

May 2026
OpenAI AI Engineer technical screen, one round focused entirely on numerical computing for ML. The problem was deceptively hard once you got past the basic softmax definition, and the streaming extension really separated people who actually understand floating point from people who just memorized formulas.
  • Given a vector of logits, compute the entropy of the softmax distribution in a numerically stable way, avoiding overflow in the exponentiation and underflow in the log.
  • Extend the entropy computation to a streaming setting where logits arrive one at a time. Maintain the entropy incrementally without reprocessing previous logits, and handle the case where a new logit exceeds the current running maximum.

“I knew the log-sum-exp trick going in, so the first part felt okay.”

View Post