Senior AI Engineer (REMOTE)

Role overview

Build and improve language models through hands-on experimentation, with a major focus on reinforcement learning. The work spans the full modeling lifecycle—from structuring specialist knowledge and preparing data to training, evaluation, and communicating results—and calls for close collaboration with subject matter experts.

Responsibilities - Elicit specialized knowledge and turn complex source material into structured, testable data; investigate quality and coverage issues with relevant experts. - Analyze data, curate training and evaluation sets, and combine automated checks with human review. - Design reinforcement learning environments, reward functions, and verifiers; train and fine-tune language models using supervised and reinforcement learning methods. - Set baselines, run controlled experiments, and investigate issues such as unstable training, reward hacking, or data leakage. - Build evaluation metrics and benchmarks, compare approaches, and track experiments with reproducible configurations and compute budgets. - Document findings and limitations, then explain results to technical teams, domain experts, and business stakeholders.

Requirements - At least three years of practical language model training or fine-tuning experience, including end-to-end reinforcement learning work designing an environment and reward function and training a policy. - Working knowledge of language-model policy optimization methods such as PPO, DPO, or GRPO, including their limitations. - Strong Python and PyTorch skills, familiarity with Hugging Face tools, and experience with parameter-efficient fine-tuning. - Experience building LLM agents or multi-turn systems and designing evaluations with held-out data and automated grading. - Ability to translate complex documentation into precise, structured specifications and to write clear technical reports. - Bachelor’s degree in a relevant field or equivalent practical experience; a graduate degree is preferred.

Nice to have

Experience in healthcare or another regulated field; handling sensitive data under privacy requirements; schema-constrained model output; knowledge graphs or ontologies; synthetic data; broader reinforcement learning methods; or relevant research and open-source contributions.

Benefits and work setup

This is a fully hands-on individual contributor position. Roughly half the work centers on reinforcement learning and model training, with the rest covering knowledge engineering, data analysis, and evaluation. Data work must follow applicable privacy rules, favoring de-identified or synthetic data where possible.