← Back to Directory

OpenAI interview notes, page 14 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
Coding round for a software engineer role at OpenAI. Single algorithmic problem, grid-based BFS. Pretty standard for this kind of company but the multi-source angle is where people slip up.
  • You're given an m x n grid where each cell is either empty, a fresh orange, or a rotten orange. Each minute, a fresh orange adjacent (in 4 directions) to a rotten one also becomes rotten. What's the minimum number of minutes until no fresh oranges remain? Return -1 if it's not possible.

“The BFS part clicked pretty fast but I almost started with a single-source approach and would've gotten the wrong answer.”

View Post

OpenAI·Software Engineer·Onsite - System Design / Architecture

Apr 2026
System design round at OpenAI for a software engineer role, and this one was heavily front-end focused which I wasn't fully expecting. The question had a lot of moving parts and I think I handled some pieces well but definitely got lost in the weeds on a few constraints.
  • Design an AI chatbot system with a front-end focus where conversation history lives only in the browser, bot responses are streamed token by token, a page refresh wipes the session, and client-side auth must not expose provider secrets.

“This question is really like five questions stitched together and I underestimated that at first.”

View Post

OpenAI·Software Engineer·Onsite - Coding / Algorithms

Apr 2026
OpenAI SWE interview with a multi-part combat simulation coding problem. Three incremental sub-questions building on each other, which is a format I hadn't prepped for specifically and it showed.
  • Implement a monster fighting / combat simulation across three incremental parts, where each part builds on the previous one.

“The three-part structure is what got me.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026
Got a graph theory problem for an OpenAI SWE round, the kind that looks deceptively clean on the surface but requires you to actually remember your Eulerian path theory from undergrad. One problem, pretty well-defined constraints, and you either know the trick or you don't.
  • Given an undirected graph where each edge must be drawn exactly once, find the minimum number of continuous strokes needed to draw all edges. A stroke can start anywhere and revisit vertices but cannot redraw an edge.

“I recognized this as an Eulerian path/trail problem pretty quickly, which was lucky.”

View Post

OpenAI·Software Engineer·Technical Phone Screen

Apr 2026
Got a system design question at OpenAI for a software engineer role, centered on building a friend recommendation feature similar to what you'd see on any major social platform. Pretty classic graph traversal problem but the scoring and complexity discussion is where it gets interesting.
  • Given a social network graph and a target user X, design a system to recommend the top K second-degree connections (friends-of-friends) that X doesn't already follow. The score for each candidate is the number of mutual first-degree connections with X. Walk through your approach including traversal strategy, scoring, ranking, and complexity analysis.

“I started with BFS up to depth 2 which felt right, but I fumbled a bit explaining why BFS over DFS here.”

View Post

OpenAI·Machine Learning Engineer·Onsite - Coding / Algorithms

Apr 2026
OpenAI MLE coding round centered on a broken Transformer implementation where you hunt down planted bugs until the training loop actually converges. There's a follow-up that's either a classifier head swap or a KV cache plug-in, and the whole thing moves fast.
  • You're given a mostly working Transformer implementation with several bugs planted in it. Find and fix all of them so that training loss converges and outputs are correct.
  • Modify the Transformer to work as a classifier: replace the language model head with a classification head, adjust the prediction and loss accordingly. Some interviewers ask for mean-pooling over the sequence before the final projection.
  • Given a skeleton KV cache class, plug caching into the attention mechanism of the Transformer, handle positional embedding indexing correctly for cached steps, and wire up the pass-through parameters.

“This is the core exercise and it's less about memorizing a bug list than about actually understanding why each thing is broken.”

View Post

OpenAI·Machine Learning Engineer·Onsite - Coding / Algorithms

Apr 2026
ML coding round at OpenAI for a machine learning engineer role. 75 minutes, one big question: implement a Transformer encoder block from scratch in PyTorch without touching any of the built-in attention modules. Brutal but fair.
  • Implement a full Transformer encoder block from scratch in PyTorch, including scaled dot-product attention, multi-head attention, a position-wise feed-forward network, residual connections with LayerNorm, and sinusoidal positional encodings. No nn.Transformer or nn.MultiheadAttention allowed. Also discuss attention complexity and how a decoder block would differ.

“This is the whole round, basically.”

View Post

OpenAI·Machine Learning Engineer·Onsite - System Design / Architecture

Apr 2026
OpenAI ML Engineer interview that was more software design than ML. The whole thing centered on building a turn-based game simulation, and the real test was how cleanly you could model it and how gracefully your design held up as they kept piling on new requirements.
  • Design and implement a two-player turn-based game simulation (like tic-tac-toe). Model the players, board state, turn logic, and win/draw conditions, and expose a clean API for starting a game, taking a turn, querying state, and detecting when the game is over.
  • How does your design handle iterative feature additions like variable board sizes, more than two players, alternative win conditions, undo/replay, and persistence? What would you refactor and what are the trade-offs?

“I went straight to the board representation and kind of forgot to think about the API surface first, which bit me later.”

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 chess game from scratch. Pretty deep dive, way more than I expected for a single question.
  • Design the object-oriented architecture for a chess game. Cover the core entities, legal move generation, special moves like castling and en passant, check/checkmate/stalemate detection, draw conditions, move history, and undo/redo functionality.

“This question has a lot more surface area than it looks.”

View Post

OpenAI·Software Engineer·Onsite - Coding / Algorithms

Apr 2026
OpenAI SWE coding round, one big design-and-implement question that took the whole session. The problem sounds straightforward until you realize extensibility is the whole point and they're watching how you structure things, not just whether it runs.
  • Design and implement an extensible chat application that supports multiple bots and multiple channels. Each channel should have independent message history and bot state. Adding new bot types should not require changes to core logic.

“This one ate the entire interview.”

View Post