Director, Model Research & Development

## Director, Model Research & Development Apply:
Hybrid:
Switzerland, Zug, Zug:
United Kingdom, London:
Full time:
Posted Today:
End Date:
October 31, 2026 (24 days left to apply):
JREQ
203994
** Job Description
** Thomson Reuters Labs is the global research and development division of Thomson Reuters. We bring together scientists, engineers, and domain experts to explore emerging technologies and develop compelling new AI products.  From research and prototyping to customer success, our work bridges frontier science with practical application. Our innovations help professionals make better decisions, faster – for example, with intelligent legal assistants, automated compliance workflows, or agentic tax engines.  

We are always looking for creative, highly adaptable scientists and engineers who enjoy solving hard problems that matter.

Thomson Reuters Labs builds and ships custom models for professional workflows requiring fiduciary-grade AI. The Thomson LLM model family was developed by training leading open-weight models on decades of authoritative Thomson Reuters content, with thousands of hours of subject-matter-expert input. Owning the model layer is how we earn trust in the most regulated, highest-stakes professional workflows: it lets us tune for the accuracy, citation-grounding, and jurisdictional awareness our customers require, set our own training and evaluation priorities, and build the unit economics that let AI scale profitably across products used by more than a million professionals.  

Reporting to the Vice President of Model Research & Development, the Director of Model Research & Development is responsible for hands-on technical direction of the research program, including the evolution of the post-training and data strategy that turns capable base models into accurate, well-calibrated, and trustworthy models for regulated use in professional workflows. You’ll work closely with the team designing and running experiments, building training and evaluation pipelines, and publishing findings at top tier conferences.

See the detailed technical report for more information on the latest model release:
Thomson 1.0 Technical Report on Hugging Face.  
** Key Responsibilities*
* * Own the hands-on execution of post-training for Thomson's LLMs: supervised fine-tuning, preference optimization (e.g., DPO), and reinforcement learning, including RL in agentic, multi-step settings where models learn to use tools.
* Stand up and run the online, agentic reinforcement-learning pipeline, training with subject-matter experts in the loop.
* Own data selection, mixture optimization, synthetic-data generation, and evaluation design, and be able to point to a specific change and its measured effect on model behavior.
* Recognize when a training run is going wrong before it finishes, and know what to do about it.
* Work day-to-day with infrastructure and evaluation teams to keep training and eval pipelines reliable.
* Bring findings and recommendations forward to inform roadmap and prioritization decisions.  
** Required Qualifications*
* * Post-training and reinforcement learning depth. Hands-on command of supervised fine-tuning, preference optimization, and reinforcement learning, including RL in agentic, multi-step settings where models learn to use tools and complete tasks.
* Data-centric model development. You've personally shaped data selection, mixture design, synthetic-data generation, and evaluation design, and can point to a specific change and its measured effect.
* Engineering depth. Hands-on command of distributed training, data pipelines, and evaluation infrastructure, able to build and debug these systems yourself, not just direct others who do.
* Track record. Shipped models, widely-used open-source…