Senior Research Scientist (LLM Post-Training)

Location: London

Company

DeepL is a global AI product and research company building secure language AI for translation, writing, and voice translation.

What you will do

• Develop steerable translation models conditioned on user preferences, rules, and context.
• Conduct hands-on post-training research for core translation models, including supervised fine-tuning, knowledge distillation, preference optimization, and translation-quality-tuned reinforcement learning.
• Build reward models and evaluator models for translation, including rubric- and reference-based grading, and investigate/mitigate reward hacking and quality-estimation failure modes.
• Advance multimodal translation by driving research toward models that ingest multimodal content and context.
• Own the full model lifecycle from prototyping and ablations to training, evaluation, optimization, and production deployment with engineering.
• Set up evaluation, reproducibility, monitoring, and continuous improvement practices; mentor researchers and engineers.

Requirements

• Proven experience making large models steerable and instruction-following using effective behavior-instillation methods (e.g., instruction tuning, latent space methods, steering vectors, constrained encoding/decoding).
• Deep hands-on expertise in LLM post-training (SFT, DPO), knowledge distillation, and/or reinforcement learning (RLHF/RLAIF, PPO/GSPO, reward modeling).
• Strong data-centric skills for synthetic-data and preference-data pipelines, including LLM-as-judge generation, data curation/filtering, and reasoning about data mixtures and ablations.
• Experience designing evaluation and reward signals using automatic metrics, LLM-as-judge evaluation, non-verifiable rewards, and human-in-the-loop evaluation.
• Hands-on builder mindset: training models, running experiments, debugging pipelines, and integrating ML systems into production.
• Strong coding and experimentation skills (Python, PyTorch/JAX/TensorFlow) and clear communication aligning research with product and engineering priorities.

Culture & Benefits

• Hybrid work schedule with office attendance twice a week.
• Flexible working hours aligned with team time zones.
• Virtual Shares for an ownership mindset.
• Regular in-person team events and monthly full-day hacking sessions (Hack Fridays).
• 30 days of annual leave (excluding public holidays) and access to mental health resources.
• Competitive benefits tailored to location.