← Back to Directory

OpenAI interview notes, page 8 of 67

Mid-sized

Most of these OpenAI write-ups come from Software Engineers (460), then Machine Learning Engineers (137) and Backend Engineers (23). The most reported rounds are the technical phone screen (335), the system design round (229), the onsite coding round (32) and the behavioral / leadership round (22).

667 interview notes · updated Jul 2026

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026
OpenAI SWE interview with a grid infection simulation problem. The problem had multiple sub-tasks ramping up in difficulty, which I wasn't fully expecting. Not a bad experience but the K-of-N neighbor counting tripped me up early.
  • Given an m x n grid where some cells start infected, and a healthy cell becomes infected each minute only if at least K of its 4-directional neighbors are currently infected, find the minute when all reachable cells are infected, or determine if some cells will never become infected.

“My first instinct was BFS and I went down that road for a few minutes before realizing the K-neighbor condition makes it way messier than standard BFS.”

View Post

OpenAI·Machine Learning Engineer·Technical Phone Screen

Apr 2026
OpenAI ML Engineer interview with a pretty gnarly debugging exercise plus a full classifier pipeline task. Not a vibe-check round at all, very hands-on from the start.
  • You're given a transformer-based model with four failing unit tests. Two bugs are documented, two are new. Find and fix all four so the model trains and evaluates correctly.
  • Given a labeled dataset, write code to train a classifier, analyze class balance and feature distributions, and report key performance metrics.

“The two known bugs were almost a trap in how easy they felt.”

View Post

OpenAI·Machine Learning Engineer·Technical Phone Screen

Apr 2026
Went through a technical screen for an ML engineer role at OpenAI. Just the one question but it was deceptively deep once you start pulling on the thread.
  • What happens when you change the context window size of a large language model?

“Started with the obvious stuff, longer context means the model can attend to more tokens, more memory, slower inference.”

View Post

OpenAI·Software Engineer·Onsite - Behavioral / Leadership

Apr 2026
Interviewed at OpenAI for a software engineering role and got hit with a broad AI safety question that was more open-ended than I expected. Less about coding, more about how you actually think about risk and responsibility at scale.
  • You're working at a company that deploys AI systems used by millions of people. How do you think about AI safety from a practical engineering and product standpoint, including the risks you'd prioritize, how you'd embed safety into the build process, and how you'd balance moving fast with deploying responsibly?

“This one sprawled in a way that was hard to contain.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026
OpenAI SWE interview focused on a KV store coding problem with some pretty interesting follow-ups that escalated fast. The AOL implementation at the end was the part I didn't fully see coming.
  • Implement serialization and deserialization for a key-value store.
  • What would you do if a single file can't hold all the data from the KV store?
  • After restoring data from disk, how would you reduce overhead when the system shuts down again?
  • Implement an append-only log (AOL) for the key-value store.

“Felt okay about the core problem.”

View Post

OpenAI·Software Engineer·Onsite - Behavioral / Leadership

Apr 2026
Behavioral round for a Research Engineer role at OpenAI. Five questions, all values-heavy, and the whole thing felt less like a skills screen and more like they were trying to figure out if you actually believe in what they're building. Not what I expected going in.
  • Why do you want to join OpenAI specifically?
  • Why are you leaving your current role, and what has driven your past job changes?
  • What is your perspective on AGI, and where do you think we are in that journey?
  • What does AI safety mean to you personally?
  • Your direct manager asks you to ship an AI feature without running safety tests. What do you do?

“I had an answer ready but it felt thin once I said it out loud.”

View Post

OpenAI·Machine Learning Engineer·Hiring Manager Screen

Apr 2026
Thirty-minute hiring manager screen for an MLE role at OpenAI. Pretty standard format but the expectation to have a polished project walkthrough AND thoughtful questions ready made it feel denser than a typical screen.
  • Walk me through a relevant past project end to end, including the problem, your role, the decisions you made, trade-offs, and what the outcome looked like in terms of metrics.
  • Tell me about a time you took ownership of something that wasn't strictly your responsibility.
  • Describe a situation where you had a conflict with a teammate or stakeholder and how you handled it.
  • How do you handle competing priorities when things are ambiguous and there's no clear direction from leadership?
  • Can you give an example of a time you had to manage up, like pushing back on a decision or getting buy-in from someone senior?

“This is the one I probably over-prepared for and still fumbled the metrics part.”

View Post

OpenAI·Machine Learning Engineer·Technical Phone Screen

Apr 2026
Coding round for an ML Engineer role at OpenAI. One meaty implementation question that looked straightforward until the edge cases started piling up.
  • Write a function that checks whether an installed version string satisfies one or more version constraint specifiers. It should handle pre-release suffixes, missing patch/minor components, and raise errors for invalid inputs. Also discuss test cases for boundary conditions and how you'd extend the solution to support PEP 440 syntax.

“Started confident, got humbled fast.”

View Post

OpenAI·Machine Learning Engineer·Technical Phone Screen

Apr 2026
OpenAI ML engineer interview with a pretty brutal debugging exercise. They handed me a broken Transformer implementation and wanted me to find and fix four distinct bugs across the architecture. No hints, just code and a timer.
  • Given a Transformer implementation with four planted bugs, find and fix the shape error in the feed-forward (MLP) block. Explain why the broken version fails during training.
  • Identify and fix the attention mask bug. The mask may be applied with the wrong shape, wrong sign, or to the wrong tensor entirely. How does this manifest during training?
  • Find the loss backpropagation bug in the code. The gradient may not be flowing correctly due to a detached tensor, premature .item() call, wrong reduction dimension, or an in-place operation breaking autograd. Fix it and verify backward pass correctness.
  • Identify the positional encoding initialization bug. Before diagnosing, clarify with the interviewer which PE variant is used (sinusoidal, RoPE, ALiBi, or learnable). Then pinpoint the specific defect and fix it.

“This one I actually caught fastest.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026
Interviewed for a software engineering role at OpenAI and got an IPv6 address validation question that seemed straightforward until it wasn't.
  • Given an IPv6 address string formatted with dot separators like 'ffff.ffff.ffff.ffff.ffff.ffff.ffff.fffa', determine whether it is valid.

“Spent the first minute just staring at the format because IPv6 normally uses colons.”

View Post