Teams at Aleph Alpha

Engineer and optimize foundation-model pre-training systems across large-scale NVIDIA GPU clusters. Responsibilities include profiling training loops, tuning distributed parallelism strategies, optimizing communication and computation, improving GPU utilization, and collaborating on hardware-efficient model architectures. The role requires expertise in Python, PyTorch, CUDA, distributed systems, NCCL or MPI, profiling tools, and modern large language model training.

Conduct research and development on LLM post-training methods to reduce hallucinations and improve grounding, uncertainty estimation, calibration, self-correction, and reliability. Build distributed multi-node GPU training pipelines, design evaluation benchmarks, and independently lead research projects from hypothesis through production implementation and technical reporting.

Lead foundation-model pre-training research across architecture, optimization, scaling, stability, and distributed systems. Design PyTorch architectures and training recipes, develop scaling laws, diagnose convergence and distributed failures, and run experiments from small proxies to multi-thousand-GPU clusters. Collaborate with compute, data, evaluation, and post-training teams to improve model quality, reliability, and efficiency.