Research Scientist, Post-Training (AI Data Startup)

A venture-backed startup supplies datasets for model training, plus evaluation tooling, to leading AI labs. It is a small team where individual researchers directly influence how future models learn. The scientist in this seat will show, through experiments, that the data really changes model behavior.

Day to day

• Plan and execute controlled supervised fine-tuning and reinforcement-learning experiments that isolate the effect of specific datasets
• Measure gains in areas such as reasoning, tool use, extended multi-step tasks and workflows tied to specific domains
• Present results to partner labs in a form that backs up the commercial conversation
• Collaborate with internal teammates to refine data quality using what the experiments reveal

Research profile

• Two to six years of post-training work on large models, including hands-on reinforcement-learning methods
• Solid grounding in LLM training and evaluation, plus strong PyTorch and distributed-training skills
• Fascination with how the structure, selection and quality of data shape model behavior
• Able to design lean experiments quickly and draw useful conclusions from noisy outcomes
• Comfortable moving between domains such as finance, software engineering and policy
• Prefers building to theorizing; a PhD is not required

On-site in the San Francisco Bay Area. Visa transfers are considered.