AI Resident

About the role

Ema is seeking an AI Resident to tackle a significant, self-contained challenge within our Agentic AI platform. This role involves taking ownership of a problem from concept to deployment, working with real production data and a senior mentor. Residents contribute directly to systems that enhance enterprise productivity.

Key facts

What you'll do

• Develop and propose a project plan for a complex AI problem.
• Design, build, and implement AI systems within the production codebase.
• Create evaluation frameworks and conduct experiments to validate solutions.
• Deploy features behind controlled gates and document outcomes, including unexpected results.
• Work on areas such as inference optimization, agent post-training, environment design, data generation, or evaluation methods.

Requirements

• Demonstrated deep understanding in machine learning or agent systems.
• Strong engineering skills, including proficiency in Python and PyTorch.
• Experience shipping code within a large-scale production environment.
• Expertise in at least one of the following: post-training methods (SFT/DPO/GRPO-family RL), reward modeling or LLM judges, agent and tool-use systems, retrieval and memory, or evaluation design.
• Ability to design and interpret statistical experiments, understanding the difference between a single data point and a statistically significant result.
• Commitment to objective measurement and willingness to invalidate a hypothesis with clear experimental evidence.

Nice to have

• Practical experience with post-training open models (e.g., TRL, veRL, OpenRLHF) or custom loops.
• Experience debugging reward-hacked machine learning runs.
• Developed or trained systems in interactive agent environments (e.g., SWE, web, or tool-use gyms).
• Background in large-scale trace analysis, data curation, or synthetic data generation.
• Experience with serving and efficiency techniques like vLLM/SGLang, distillation, or quantization.
• Familiarity with multi-node GPU training or the infrastructure to quickly learn it.
• Publications, open-source contributions, or written work showcasing your thought process.
• Understanding of security concerns in AI, such as prompt injection, data governance, and the need for safeguards in self-improving agents.

Skills & tools

• Python
• PyTorch

Practical notes

• This is an on-site/hybrid role in the San Francisco Bay Area.
• Flexible start dates are available.
• Compensation may vary based on location, experience, and other factors.
• Ema is an equal opportunity employer.