Director, Model Research & Development

Overview

In this leadership role, you steer the post-training research program for Thomson’s LLMs, shaping data strategies, training, and evaluation to deliver accurate, well-calibrated models for regulated use. You collaborate with experimental, infrastructure, and evaluation teams to turn research into production-ready AI that supports professional decision-making. The role offers influence over roadmap priorities and exposure to leading AI methods and open-source contributions. You will publish findings and drive impactful, real-world AI applications in high-stakes domains.

Pay / Benefits

• hybrid work model
• flexible vacation
• Headspace app access
• tuition reimbursement
• volunteer days
• wellbeing resources

Responsibilities

• Oversee hands-on post-training for Thomson’s LLMs: supervised fine-tuning, preference optimization (e.g., DPO), and reinforcement learning including agentic, multi-step settings with tool use.
• Establish and operate online, agentic reinforcement-learning pipelines with subject-matter experts in the loop.
• Own data selection, mixture optimization, synthetic-data generation, and evaluation design with measurable effects linked to model behavior.
• Detect training issues early and determine corrective actions.
• Collaborate daily with infrastructure and evaluation teams to maintain reliable training and evaluation pipelines.
• Provide actionable findings to inform roadmap and prioritization decisions.

Key requirements

• Post-training and reinforcement learning depth with hands-on experience in supervised fine-tuning, preference optimization, and RL in agentic, multi-step contexts.
• Data-centric model development with demonstrated data selection, mixture design, synthetic-data generation, and evaluation design.
• Engineering depth in distributed training, data pipelines, and evaluation infrastructure; ability to build and debug systems.
• Track record via shipped models, open-source contributions, or peer-reviewed publications at top venues (e.g., NeurIPS, ICML, ACL, EMNLP).
• Technical leadership experience heading a focused training or evaluation team.

• technical leadership
• cross-functional collaboration
• strong communication
• distributed training
• data pipelines
• evaluation infrastructure

…