OpenAI Research Engineer
interview guide
Prepare for ML implementation, unfamiliar code and research-system design, with guidance on choosing practice for the team and rounds in your invitation.
What to expect at OpenAI
OpenAI research engineering interviews combine software implementation with questions about experiments, numerical calculations and the systems that support research. Tasks can involve extending a small program, debugging transformer code or explaining how you would evaluate a model. Project discussions explore the evidence behind your decisions and the work you personally contributed.
General coding, ML coding and research-scientist exercises are different formats. The guide separates them so you can prepare for your confirmed rounds.
Debugging unfamiliar software
Non-ML debugging tasks include a Python job scheduler covering races, deadlocks, lock contention, rate limiting and tests, and a Django rate limiter in a round labeled general coding. Trace the code relevant to the failure before exploring the wider application.
A useful rehearsal is to reproduce one failure, follow the relevant path and test a small fix. Do not restrict preparation to a memorized list of transformer mistakes.
Inside the full guide
8 sectionsThe process and its variants
How technical screens and later rounds differ across research engineering teams.
Progressive systems coding
Stateful implementations, changing requirements and reasoning about the supplied interfaces.
ML coding and numerical reasoning
Calculations, array shapes and explaining what an implementation is intended to compute.
Debugging transformers and unfamiliar software
Tracing failures in model code and ordinary software without relying on a memorised bug list.
Research-system design
Evaluation, retrieval and data problems where the research objective determines the design.
Mathematical depth and adjacent research tracks
Probability and research-scientist examples, with their different role requirements identified.
Project depth and motivation
Experiments, evidence, personal contribution and the decisions behind a research project.
Choose practice for your actual rounds
How to balance implementation, debugging, design and project rehearsal.