Research Scientist
Hybrid (San Francisco, CA)$150k - $250k + equityFull-timePosted 6d ago
About the role
Develop new methods for building specialized models that learn reliably and solve difficult tasks. Roles span post-training, reinforcement learning, continual learning, multimodal models, evaluation, and interpretability. Responsibilities include formulating hypotheses, running experiments, evaluating against baselines, and communicating results with limitations.
Responsibilities
- Develop new methods for building specialized models that learn reliably and solve difficult tasks.
- Work across post-training, reinforcement learning, continual learning, multimodal models, evaluation, and interpretability.
- Formulate hypotheses, implement and run experiments, compare against meaningful baselines, and collaborate with research engineers to turn findings into practical capabilities.
- Communicate results clearly, including limitations and competing explanations.
Qualifications
- Typically 3+ years of relevant ML research experience, or equivalent demonstrated research contributions
- Independent drive for research with ability to discuss contemporary research in language model training
- At least one substantial original research contribution demonstrated through a paper or open-source implementation
- Depth in at least one of post-training, reinforcement learning, continual learning, multimodal learning, evaluation, interpretability, systems, or inference
- For work focused on model training: typically 1+ year of hands-on language or multimodal model training, including doctoral research, or equivalent completed training projects
Skills
Required:ML researchPost-training learningEvaluation
Nice to have:Reinforcement learningContinual learningMultimodal learningInterpretability
Last updated Oct 21, 2026. Interested? Apply on the company site.
About TrainLoop
Reasoning Fine-Tuning
TrainLoop makes it effortless for developers to supercharge LLM performance through reinforcement learning.
developer-toolsgenerative-aireinforcement-learning
More open roles at TrainLoop
- Research Engineer6d agoHybrid (San Francisco, CA)
- Head of Enterprise6d agoHybrid (San Francisco, CA) · Sales
- UX / Product Designer6d agoHybrid (San Francisco, CA)