Senior Research Scientist FMTA

We are the team behind DeepL’s post-training stack for large language models. We focus on developing algorithms and systems that align pre-trained models with tasks and performance goals through techniques like reinforcement learning. As a Senior Research Scientist, you’ll design, implement, and deploy cutting-edge research in reinforcement learning and post-training at scale, driving innovations that make it into production.

Responsibilities

Build and deploy state-of-the-art reinforcement learning pipelines at scale; post-train large (multi-modal) models to align them with human intent and enable general capabilities such as reasoning; keep the entire lifecycle of research and production in mind from idea conception, theoretical modeling, prototyping, ablation studies, to production deployment; build and foster external collaborations with academic and industrial partners; follow scientific and technical standards for experimentation, reproducibility, and model evaluation; collaborate deeply with Engineering, ML Platform, and HPC teams to deliver robust and reliable model updates to users

Requirements

Deep technical background; strong leadership skills; proven track record of driving research in reinforcement learning or large-scale model alignment to production; strong practical background; creative mindset; passion for solving hard problems with real-world impact; solid mathematical background; masters degree, diploma, PhD, or equivalent industry experience in mathematics, physics, computer science, or a related field; deep practical experience in Python; experience with at least one modern machine learning framework such as PyTorch, TensorFlow, or JAX; track record of leading self-directed research projects that deliver tangible results

Nice to have

Experience working with large compute clusters and ML infrastructure is a plus; expertise in deep reinforcement learning (RLHF/RLAIF/RLVR) is a plus; hands-on experience scaling and deploying LLMs or other foundation models in real-world systems is a plus

Benefits

Diverse and internationally distributed team; open communication, regular feedback; hybrid work, flexible hours; monthly full-day hacking sessions; 30 days of annual leave; access to mental health resources; competitive benefits