OpenAI Interview Questions — Page 10

OpenAI·Machine Learning Engineer·Technical Phone Screen

Apr 2026

Technical screen for an MLE role at OpenAI, focused entirely on backprop through a chain of matrix multiplications with no nonlinearities. Pretty deep dive, more like a math exam than a coding round. Left feeling like I'd done okay on the main derivation but fumbled some of the follow-ups.

Derive the gradient of the loss with respect to each weight matrix W_j in a block that computes Z = W_1 * W_2 * ... * W_n * X, where there is no nonlinearity inside the block. Show why the formula takes the form it does, not just what the formula is.
Verify that the gradient you derived for W_j actually has the same shape as W_j itself, and show how the batch dimension gets contracted away.
Describe an efficient implementation of the backward pass that computes all n gradients in O(n) matrix products instead of O(n^2). How do you avoid recomputing overlapping prefix and suffix products?
How does the backward pass change if you also need the gradient with respect to the input X?
If two of the matrices in the chain are tied (the same parameter appears at two positions j and k), how does the gradient for that shared parameter change?
With a large number of matrices in the chain, what numerical issues arise from repeated matrix products and how would you address them?

“This was the core question and it took up most of the session.”

View Post

OpenAI·Software Engineer·Onsite - System Design / Architecture

Apr 2026

OpenAI system design round for a software engineer role. The prompt came pre-loaded with API and functional requirements, so most of the session was spent on fault tolerance and worker reliability, which is where things got interesting.

Design a video generation system similar to Sora, with a focus on handling worker failures and crashes during long-running generation jobs.

“The API and functional requirements were already written out in the prompt, so I didn't have to spend time scoping.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026

Went through a technical phone screen for a software engineering role at OpenAI. The problem was a grid simulation with a layered death rule on top of standard BFS infection spread. Tricky enough that I had to think carefully about what state to track per cell.

You have an N×M grid where some cells start infected. Infection spreads to 4-directional neighbors each day. Additionally, an infected cell dies D days after infection if it ever had K or more simultaneously-infected neighbors during those D days. How many plants die in total?

“The base BFS part was fine, I've done enough flood-fill stuff to get through that quickly.”

View Post

OpenAI·Software Engineer·Onsite - System Design / Architecture

Apr 2026

System design round at OpenAI for a software engineer role, focused entirely on LLM inference serving at scale. One long, dense question with a lot of moving parts. The kind of interview where you realize halfway through that you've been talking about the wrong layer.

Design an online inference serving system for a GPT-style large language model. The system needs to handle around 100K concurrent requests across a GPU fleet with near real-time latency constraints. Cover the API and request data flow, whether requests should be persisted, failure handling and retry logic, how to avoid duplicate inference when batches overlap, and dynamic batching on GPU. Also discuss which components are actually necessary given the latency budget versus what you'd skip.

“This one sprawled.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026

OpenAI software engineer interview that eventually pushed into some pretty deep territory. The rare follow-up here was implementing the async message dispatch logic from scratch rather than just assuming it exists, which I did not see coming.

Implement the sendAsyncMessage function itself, including the async dispatch logic, and write test cases that exercise the full system end to end.

“Most of the session I assumed this utility was just given to me.”

View Post

OpenAI·Software Engineer·Onsite - System Design / Architecture

Apr 2026

System design round at OpenAI for a software engineer role, centered entirely on designing a Sora-style text-to-video generation platform. Brutal scope, lots of moving parts, and the safety angle caught me more off guard than I expected.

Design the backend platform for a text-to-video generation product where users submit a natural-language prompt and receive a generated video clip. The model itself is a black box. Focus on everything around it: request handling, queuing, GPU-backed generation, safety, storage, and delivery at scale.
How would you define the client-facing API and the full lifecycle of a generation job, from the moment a user submits a prompt to the point where they receive the finished video?
How would you design the GPU worker pipeline and scheduling system to keep an expensive GPU fleet highly utilized while processing a mix of job types with varying durations?
How would you implement safety checks for both the input prompt and the generated video output, and where in the pipeline do these checks live?
How would you handle storing and delivering large video files to users efficiently and economically?
How would you support priority tiers for free versus paid users without indefinitely starving free users of GPU access?
A generation job has been running for 4 minutes and the GPU worker crashes. Walk through exactly what happens and how the system recovers.
How would you add an iterative editing feature, like extending a clip or remixing a segment, on top of the existing one-shot generation design?
How would you safely roll out a new, more expensive model version and A/B test it without blowing the GPU budget?

“This is a massive question and I think I spent too long on the API layer before getting to the GPU scheduling piece, which is clearly where the real complexity lives.”

View Post

OpenAI·Machine Learning Engineer·Technical Phone Screen

Apr 2026

OpenAI ML engineer round, one deep question that sounds like a fun puzzle until you're actually in it. The 1-NN as a neural net thing is the kind of problem where you think you see the answer immediately and then realize you're missing half the pieces.

Express the 1-Nearest-Neighbor algorithm as a pure neural network forward pass (no training, inference only), where every operation is a linear layer followed by an activation. The network should take a query vector and output the nearest training example's label.

“I knew the 1-NN problem from a prior round so I wasn't starting from zero, but rephrasing it as a literal forward pass with Wx+b layers and at least one activation was a different beast.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026

OpenAI Research Engineer interview with a meaty system design problem around social network data structures. The question kept evolving mid-interview which was a bit stressful but also kind of interesting.

Design a social network data structure that supports an update operation (user A follows user B at timestamp t) and a check operation (does A follow B at a given time t)?
Extend the design to support mutual-follower (bidirectional) queries efficiently.
Further extend the design to recommend users for A based on second-degree connections, and rank recommendations by how many distinct intermediaries connect A to the candidate.

“I went with a map from user pairs to a sorted list of timestamps.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026

OpenAI software engineer coding round, one problem with multiple sub-parts centered on epidemic/disease spread simulation. Got through three of the sub-problems before the interviewer cut it off, which stung a little since there was still time left on the clock.

Simulate the spread of an infectious disease through a population. The problem had multiple sub-parts building on each other.

“I'd actually prepped for this general topic so it wasn't a total blindside.”

View Post

OpenAI·Software Engineer·Onsite - System Design / Architecture

Apr 2026

System design round at OpenAI for a software engineering role, and they threw a full Xiangqi game implementation at me. Not what I expected but in hindsight it makes sense given the complexity they're after.

Design and implement an online Xiangqi (Chinese Chess) game, covering board representation, piece movement rules, move validation, game state management, and a basic networking layer for multiplayer.

“This was a lot.”

View Post