Post-Training ML Research Scientist (RL & Data) - SF On-site

True Talent Solutions Partners; San Francisco (CA); Presencial; USD 150.000 - 250.000; Jornada completa

Descripción de la vacante

True Talent Solutions Partners is seeking an experienced ML researcher to join a post-training R&D team in San Francisco. You will design controlled experiments to quantify how data affects model capabilities, generalization, and alignment, and translate findings into practical recommendations.

You will publish research and build evaluation infrastructure, while collaborating with partner labs to ensure reproducible results across data and model configurations.

Formación

• Minimum 3 years of machine learning research with post-training or RL exposure.
• Experience implementing RL or post-training techniques in research or production.
• Strong results in published research or shipped ML systems.
• Deep familiarity with large language models and transformer architectures.
• Proficiency in Python and major DL frameworks; distributed training experience.

Responsabilidades

• Design and run supervised fine-tuning and RL experiments to test data impacts.
• Build evaluation infrastructure, benchmarks, and new data types.
• Publish external research, blogs, and technical reports.
• Collaborate with internal teams to improve data quality from experiments.
• Work with partner labs to produce reproducible evidence about data and models.

Conocimientos

ML research

Post-training RL

Python

PyTorch

TensorFlow

Distributed training

Open-source contributions

Educación

PhD in ML/CS or equivalent

Herramientas

PyTorch

TensorFlow

Descripción del empleo

True Talent Solutions Partners is seeking an experienced ML researcher to join a post-training R&D team in San Francisco. You will design controlled experiments to quantify how data affects model capabilities, generalization, and alignment, and translate findings into practical recommendations.

You will publish research and build evaluation infrastructure, while collaborating with partner labs to ensure reproducible results across data and model configurations.