A frontier data provider equipping foundational AI labs with datasets that shape the next generation of models.
Research Scientist - Post-Training
We're building a new research function inside a well-funded, 40-person company that supplies the data layer for the most advanced AI systems in the world. This team operates at the intersection of empirical science and model improvement—designing post-training pipelines that measurably lift reasoning, instruction-following, and alignment. You'll own experiments from hypothesis through execution, working closely with a small group of researchers who treat post-training as a first-class research problem, not an afterthought. The outputs of your work will land directly in frontier models trained by our partners, making rigor and reproducibility non-negotiable.
What they're looking for
- 1–4 years of hands-on experience training or fine-tuning large models, with clear ownership of at least one post-training research project (RLHF, DPO, instruction tuning, or similar).
- Strong fluency in Python and PyTorch, including experience writing clean, modular experiment code and debugging training pipelines.
- Demonstrated ability to formulate research hypotheses, design careful ablations, and draw conclusions that hold up under scrutiny.
- Comfortable reasoning about data quality, annotation design, and how dataset composition interacts with post-training objectives.
- Preference for working in a high-trust, low-process research environment where you ship experiments frequently and learn from failure quickly.