AI Researcher - Pre-training (f/m/d)
Core
Design and run experiments on the architecture, optimization, and scaling of large-scale foundation model pre-training.
Role type
Senior IC AI Researcher (Pre-training)
Builds
Next-generation foundation models (e.g., Kolibri-1) for high-stakes domains like finance and public administration.
Domain
Artificial Intelligence / Large Language Models / Pre-training
Deliverable
production ML models
Required skills
Python, PyTorch, Machine Learning fundamentals, Mathematical reasoning (linear algebra, probability, optimization), Transformer architecture knowledge, Large-scale pre-training experience, Experimental design, Code reproducibility
Preferred skills
Distributed training (multi-GPU/multi-node), Scaling laws, Hyperparameter transfer, Efficient architectures (MoE), Top-tier research publications, Systems curiosity
Technologies
PyTorch, NVIDIA Blackwell GPUs, Megatron-LM, DeepSpeed, torchtitan
Responsibilities
Design and run ablation experiments on training recipes and architecture; Analyze scaling laws and predict behavior at multi-thousand-GPU scale; Monitor large-scale training runs and debug convergence issues; Reproduce literature results and bring ideas to the team; Collaborate with Compute, Data, and Evaluation teams.
Seniority
Senior, hands-on IC
