Senior Research Scientist FMTA

Overview

In this role, you will lead cutting-edge reinforcement learning research and post-training at scale to improve DeepL’s multi-modal models. You’ll shape how models learn beyond pre-training, aiming for safer, more capable systems used in production. You’ll collaborate with cross-functional teams and external partners to advance the foundation model stack. This is a research-driven position at the Foundation Model Task Adaptation team, offering impact through real-world deployment and scalable experiments. Join us to help push the boundaries of AI alignment and practical AI deployment at global scale.

Pay / Benefits

• Hybrid work
• 30 days annual leave
• Hack Fridays monthly
• Diverse, international team
• Mental health resources
• Competitive benefits tailored to location

Responsibilities

• Design, implement, and deploy reinforcement learning research and post-training pipelines at scale
• Post-train large multi-modal models to align with human intent and enhance capabilities, safety, and efficiency
• Manage end-to-end research lifecycle from ideation to production deployment
• Foster external collaborations with academic and industrial partners
• Maintain rigorous experimentation, reproducibility, and model evaluation standards
• Collaborate with Engineering, ML Platform, and HPC teams to deliver robust model updates to users

Key requirements

• Deep technical background in reinforcement learning or large-scale model alignment to production
• Strong leadership and track record of self-directed research delivering tangible results
• Solid mathematical background; advanced degree or equivalent in relevant field
• Proficiency in Python and at least one ML framework (PyTorch, TensorFlow, or JAX)
• Experience with large compute clusters and ML infrastructure a plus
• Experience scaling and deploying LLMs or foundation models in real-world systems a plus
• Expertise in RLHF/RLAIF/RLVR is a plus

• Leadership
• Creative problem solving
• Strong collaboration and communication
• Python
• PyTorch
• TensorFlow

…