The role
- COMP
- $200K - $450K
- EQUITY
- Competitive equity
- LOCATION
- San Francisco
- WORKPLACE
- Hybrid
- EXPERIENCE
- 2 - 8 years
- VISA
- None, Visa transfers
- STACK
- Python, PyTorch
- INDUSTRY
- AI
/roles — ROLE_339
Machine Learning Research Scientist
Seed-stage lab training reasoning models and agents that automate the machine learning research process
The company
Early-stage AI lab building autonomous agents that carry out machine learning research on their own, sold commercially to enterprises that train their own models.
- STAGE
- Series A-stage
- FUNDING
- $10M+ raised
- TEAM
- ~10 people
- FOUNDED
- 2024
- BACKING
- VC-backed
JD — the work
About the role
The company builds AI systems that carry out AI research on their own, and this role sits at the center of that work. In close partnership with the founder, you would develop research agents that generate ideas, run experiments, and improve customer models, then help engineering turn them into production systems. The focus is RL post-training and fine-tuning of reasoning models, not wiring together LLM APIs. The role is based in the Bay Area.
What you'll do
- Design, with the founder, agents that generate hypotheses, run experiments, and lift customer model performance
- Apply RL post-training and fine-tuning to reasoning models so they can automate parts of ML research
- Partner with engineers to ship research systems that hold up in production
- Keep up with new results in automated research and machine learning, bringing promising ideas back
- Pursue new, unproven research directions with a high degree of independence
What they're looking for
- A PhD in CS, ML, AI, or an adjacent field, or a comparable research record without the degree
- Publications at top ML conferences, or comparable output from a corporate AI research lab
- Hands-on depth in model training, such as RL, deep neural networks, or evolutionary methods
- Model-training experience, not just multi-agent apps on hosted LLM APIs or retrieval-based agents
- Real drive to speed up scientific discovery with AI
Nice to have
- Distributed training at scale (more than 64 GPUs) or production ML pipeline experience
- Background in research automation or AI for science