Senior LLM Engineer

3 days ago

Barcelona, SpainSenior

Responsibilities

Train and post-train LLMs and multimodal models on multi-node GPU clusters using supervised fine-tuning, preference optimization, reinforcement learning, and distillation.
Design model compression and healing techniques, including quantization, pruning, quantization-aware training, and tensor-network methods.
Build reinforcement-learning environments, reward functions, and verifiers for reasoning, coding, agentic, world-model, and physical-AI tasks.
Research and prototype post-transformer and hybrid architectures, including state-space models, linear attention, mixture-of-experts, and diffusion language models.
Build large-scale data pipelines for curation, filtering, deduplication, and synthetic data generation.
Design ablations and evaluations across reasoning, coding, long-context, instruction-following, and agentic benchmarks.
Optimize model inference across accelerators and edge chips by profiling latency, memory, and throughput.
Document experiments in internal reports and, where relevant, papers and open-source releases; review code, mentor teammates, and incorporate current research.

Master’s or PhD in Computer Science, AI, Mathematics, Physics, or a related field, or equivalent industry experience.
At least 4 years of hands-on experience training or post-training transformer models.
Strong PyTorch skills and comfort working below the library level with model code, custom losses, and parallelism configuration.
Experience with distributed training frameworks such as FSDP, DeepSpeed, or Megatron-LM on multi-node clusters using Slurm or Kubernetes.
Hands-on post-training experience with SFT, DPO, and at least one reinforcement-learning method such as PPO or GRPO.
Strong foundations in linear algebra, optimization, probability, and training dynamics.
Understanding of GPU memory hierarchy, mixed precision, profiling, and throughput.
Experience with rigorous ablation design, reproducibility, experiment tracking, and noise-aware evaluation.
Strong debugging and communication skills and fluent English.
Preferred qualifications include experience with NVIDIA LLM and RL tooling, rollout engines, asynchronous RL, reward models, verifiers, compression methods, post-transformer architectures, physical AI, multimodal training, alternative hardware inference, large-scale data processing, evaluation tools, publications, open-source contributions, released Hugging Face models, or Spanish.

201-500 employees