June 2026 archive · For research and skill exploration. Current job availability is unverified.
OpenAI
Software Engineer, Collective Communication
Design and implement custom networking collectives for large-scale AI model training on supercomputers.
OpenAIText summary from the June 2026 archive. original source →
The role
This role sits within OpenAI's Workload Networking team, which builds the collective communication infrastructure powering their largest training jobs. The engineer will develop performance-critical networking code in C++ and CUDA, collaborating with ML researchers to optimize distributed training across custom supercomputers.
What you'd do
- Design and implement efficient collective operations in C++ and CUDA for ML researchers
- Optimize largest training jobs to leverage different network transports on supercomputers
- Develop network simulations to guide future supercomputer architecture decisions
- Integrate custom networking collectives tightly with the training platform
What they're looking for
- Background in low-level, performance-critical software development
- Comfortable writing performance-sensitive CPU and GPU code
- Experience with distributed algorithms using RDMA
- Familiarity with network simulation techniques
Nice to have
- Prior experience with collective communication systems
Summary written by RoleDeck from the original posting. This is an extracted, own-words summary and may contain errors. The original source may have changed or expired.