AI Researcher - Pre-training (f/m/d)

Highlights

• Advance next-generation foundation models like Kolibri-1 built in Heidelberg.
• Run architecture and scaling experiments feeding directly into production training.
• Monitor large-scale runs and debug convergence, stability, and scaling behavior.

Ideal candidate

A curious ML researcher with strong fundamentals who learns from failed runs and wants to go deep on how large models learn, from proxy experiments to production scale.

Requirements

• Proficient in Python with hands-on PyTorch experience
• Strong ML foundation via PhD, strong MSc thesis, publications, open-source or substantial projects
• Solid background in linear algebra, probability, and optimization
• Understands transformers and has pre-trained LLMs from scratch, even at small scale
• Designs careful experiments and honestly reports noisy and negative results
• Fast learner with track record of going deep in unfamiliar technical areas
• Writes clean, maintainable code with focus on reproducibility
• Communicates clearly, raises problems early, and acts on feedback