Research Intern

About the Role

The Agents team investigates how to build, align, and scale frontier AI systems that can tackle complex, multi-step tasks and workflows across text and speech, with a particular focus on agentic and scientific domains. Our work sits at the intersection of agent capabilities, human-computer interaction, and infrastructure—from designing post-training methods for agentic behavior to developing evaluation frameworks for open-ended tasks where traditional metrics fall short. 

As a research intern, you will work on problems at the frontier of agentic AI, where challenges in alignment, reliability, and scalability are deeply intertwined. Projects may involve developing new training recipes for self-learning and long-horizon reasoning, curating datasets for non-deterministic scientific and agentic tasks, studying failure modes in agentic behavior, or building infrastructure that enables agent operations at scale. You'll operate in a space where algorithmic innovation, dataset and interaction design, and systems work come together to push the boundaries of what AI agents can reliably accomplish.

Responsibilities

• Research and implement novel techniques in one or more of our focus areas
• Design and conduct rigorous experiments to validate hypotheses
• Document findings in scientific publications and blog posts
• Communicate the plans, progress, and results of projects to the broader team

Requirements

• Currently pursuing a Ph.D. degree in Computer Science, Electrical Engineering, Information Science, or a related field
• Publications at leading ML, NLP, or speech conferences or journals (such as NeurIPS, ICML, ICLR, *ACL, EMNLP, Interspeech)
• Strong knowledge of Machine Learning and Deep Learning fundamentals
• Experience with deep learning frameworks (PyTorch, JAX, etc.)
• Understanding of how LLMs work
• Strong programming skills in Python
• Familiarity with Transformer architectures and recent developments in foundation models

Example Research Directions

• Training, developing, and evaluating frontier models, especially in the domain of agentic tasks and workflows
• Designing and curating datasets for frontier agents alignment and post-training
• Studying failure modes and developing safety paradigms for agentic behavior
• Research new recipes (for RL or test time scaling) for self-learning and long-context tasks completion
• Building agents that act on spoken input to carry out complex, multi-step tasks