About this role
About the Role
Work alongside an AI research team to turn experimental code into production-ready systems. You will build and improve shared post-training infrastructure that supports research data generation, processing, and evaluation.
What You'll Do
- Turn research prototypes into tested, reusable data generation, training, and evaluation pipelines.
- Build distributed experiment support, including data loading, checkpointing, and collection of agent interactions.
- Profile workloads and improve GPU utilization, memory efficiency, and data throughput.
- Develop tests, experiment tracking, and debugging tools while preserving the integrity of research results.
- Collaborate with researchers to translate ideas into maintainable software.
What We're Looking For
- Experience building ML training, inference, or data pipelines in Python.
- Strong Python skills and hands-on experience with PyTorch, JAX, or comparable machine learning frameworks.
- Understanding of machine learning experiments, including how data, numerical precision, and implementation choices affect results.
- Sound software engineering practices, including testing, profiling, version control, and documentation.
- Distributed training or data processing experience with tools such as PyTorch Distributed, DeepSpeed, or Ray is valuable.
- Familiarity with Hugging Face Transformers, vLLM, or experiment tracking platforms such as Weights & Biases or MLflow is useful.
- Experience with reinforcement learning systems or multimodal datasets is a plus.
Compensation & Benefits
Competitive compensation and equity are offered. Visa sponsorship is available.
Location
On-site in Zürich, Switzerland.