Distributed Training Engineer

About the role

Soniox is building world‑class speech and multimodal AI systems and needs a Distributed Training Engineer to scale model training to the limits of modern hardware. You will work closely with researchers to design, optimize, and maintain large‑scale training infrastructure that drives rapid experimentation across speech, video, and language models.

Key responsibilities

• Collaborate with researchers to create system‑efficient training loops for cutting‑edge speech and video models.
• Profile and optimize distributed training workloads on internal infrastructure.
• Enhance the in‑house training framework to boost hardware utilization and throughput.
• Apply algorithmic and systems‑level techniques to improve scalability, reproducibility, and stability of large‑scale training.
• Identify and resolve performance bottlenecks across the ML stack, from data loaders to kernel fusion and network latency.

Required profile

• Experience with large‑scale or multi‑modal machine‑learning pipelines.
• Enjoy diving into deep‑learning framework internals and distributed systems to make them faster and more reliable.
• Strong software‑engineering practices and a passion for correctness, reproducibility, and craftsmanship.

Required skills

• Proficient in Python.
• Knowledge of GPU programming and optimization (CUDA, Triton, XLA, custom ops).
• Understanding of training kernels and how to optimize them.
• Familiarity with distributed training frameworks and large‑scale ML infrastructure.

What we offer

• Opportunity to shape one of the most technically advanced AI platforms globally.
• Collaboration with a world‑class team of engineers and researchers.
• Influence over company growth, customer success, and AI impact on human communication.