AI Research Scientist, New Grad
Snowflake
About the Role
Snowflake is powering the era of the agentic enterprise, seeking AI-native thinkers energized by the opportunity to reinvent how work is done. This role is at the intersection of agentic AI and reinforcement learning, where research will directly shape how enterprises leverage intelligent automation. The AI Research team is pushing the frontier of autonomous, self-improving AI systems, building agents that reason, code, and learn at scale inside the Snowflake Data Cloud.
Key Responsibilities
- Design and develop agentic frameworks powered by recursive self-improvement loops.
- Build and evaluate auto research agents.
- Develop coding agents that understand, generate, and debug code.
- Conduct research in reinforcement learning with a focus on RLHF, DPO, and PPO.
- Contribute to multi-agent systems where specialized agents collaborate.
- Develop and curate training data pipelines.
- Publish research findings at top-tier venues.
Requirements & Tech Stack
- Proficiency in Python.
- Proficiency in at least one deep learning framework (PyTorch or JAX strongly preferred).
- Strong mathematical and analytical foundation.
Qualifications
- PhD in Computer Science, Machine Learning, Artificial Intelligence, or a closely related field (completing or recently completed; or equivalent research experience).
- Foundational expertise in reinforcement learning algorithms, including RLHF, DPO, PPO, or multi-agent systems.
- Research experience in LLM post-training, fine-tuning, or reasoning model development.
- Demonstrated ability to implement and experiment with agentic architectures — including tool-use, planning, and self-correction loops.
- At least one first-author or co-authored publication or preprint in a relevant AI/ML area.
Bonus Points
- Hands-on experience building or evaluating coding agents or auto research agents.
- Familiarity with recursive self-improvement frameworks or automated AI scientist paradigms.
- Experience with large-scale distributed training or efficient training paradigms.
- Background in mathematical reasoning, structured decision-making, or program synthesis.
- Exposure to domain-specific AI applications in healthcare, finance, or enterprise workflows.