Research, Mid Training

What you'll do

• Source, curate, synthesize, and improve training datasets; build filtering, verification, and quality-scoring pipelines.
• Develop and measure mid-training data mixes, behaviors, and training recipes, coordinating with post-training researchers.
• Iterate on evaluations, debug training outcomes, and study performance scaling with dataset size and new data types.

What they ask for

• Proficiency in Python and familiarity with deep learning frameworks such as PyTorch, TensorFlow, or JAX; able to debug distributed training and write scalable code.
• Bachelor’s degree or equivalent experience in Computer Science, Machine Learning, Physics, Mathematics, or a related discipline, with strong theoretical and empirical grounding.
• Clear communication and ability to explain complex technical concepts in writing.

Nice to have

• Strong grasp of probability, statistics, and ML fundamentals.
• Experience with large-model training datasets, synthetic data pipelines, knowledge injection, continued pre-training, domain adaptation, or model-based data-quality systems.

Pay

$350,000- $475,000 USD (annual base salary)

Tools

Python, PyTorch, TensorFlow, JAX

AI focus

Mid-training frontier models to establish reasoning foundations and improve knowledge, behaviors, and capabilities; includes synthetic-data generation, quality classifiers and LLM-based judges, and evaluation.