Senior AI Researcher, Reinforcement Learning

Senior AI Researcher, Reinforcement Learning at Lumen Dynamics
Senior AI Researcher- Reinforcement learning (f/m/d)
Department: Research
Location: Heidelberg
Employment Type: FullTime
Our Mission
Lumen Dynamics is one of the few companies in Europe with end-to-end in-house model development including pre- and post-training. We’re building models that have general-purpose capabilities, but also specifically excel at addressing the needs of our customers.
We're growing our post-training team in Heidelberg (or hybrid in Germany) and are looking for an AI Researcher who combines a deep theoretical understanding of reinforcement learning methods with a desire to improve on the state of the art and improve model capabilities in large-scale training.
Team Culture
At Lumen Dynamics, we foster a culture built on ownership, autonomy, and empowerment. Teams and individual contributors are trusted to take responsibility for their work and drive meaningful impact. We maintain a flat organizational structure with efficient, supportive management that enables quick decision‑making, open communication, and a strong sense of shared purpose.
About the role
As a (senior) AI Researcher for reinforcement learning you will shape and improve the underlying RL methodology, maintain a high-quality training code-base, and conduct large-scale experiments to hill-climb our performance benchmarks. This role is for you if you both have a strong theoretical background on RL and the engineering drive to bring these methods into production and improve on the methods as part of the reinforcement learning team.
In your day-to-day you will conduct large-scale reinforcement learning experiments, derive hypotheses from the results, and iterate on both the implementation and methodology based on the observations. Together with a collaborative team, you will have direct impact on the models that we ship to our customers.
This role is for Lumen Dynamics Research GmbH.
Your Responsibilities
– Hill-climb in large-scale training: Conduct large-scale LLM training runs, analyze evaluation scores in depth, propose hypotheses for improvement and directly implement them in order to maximize performance on our benchmarks.
– Theoretical innovation: Stay at the bleeding edge of RL research. You will identify, implement, and iterate on novel approaches to multi-turn reinforcement learning.
– Scale our training infrastructure: Identify bottlenecks in our training setup and optimize our RL training loops for large-scale training.
– Cross-functional collaboration: Partner with our other post-training teams to turn raw feedback into actionable training signals, ensuring that our RL iterations lead to measurable improvements in downstream performance.
Your Profile
Basic Qualifications
– A deep understanding of Reinforcement Learning theory and how it relates to modern RL methods.
– Experience with multi-node LLM training (ideally using RL). You understand how to scale multi-node RL trainings and can reason about and implement distributed algorithms.
– Familiarity with statistical methods for evaluation and experiment design.
– Ability to reason about what an evaluation/environment measures and whether it matters - not just run benchmarks, but understand them.
– Strong Python skills and comfort with ML tooling (especially torch distributed)
– Willingness to relocate to Heidelberg or travel regularly (potentially weekly).
Preferred Qualifications
– PhD in reinforcement learning or equivalent research experience.
– A history of contributions to top-tier venues (NeurIPS, ICML, ICLR, etc.) specifically regarding RL.
– Experience evaluating LLM models and crafting environments for training.
Compensation and Benefits
– Become part of an AI revolution!
– 30 days of paid vacation
– Access to a variety of fitness & wellness offerings via Wellhub
– Mental health support through nilo.health
– Substantially subsidized company pension plan for your future security
– Subsidized Germany-wide transportation ticket
– Budget for additional technical equipment
– Flexible working hours for better work-life balance and hybrid working model
– Virtual Stock Option Plan
– JobRad® Bike Lease