Senior Software Developer: Models Team (Token Factory)

Nebius is leading a new era in cloud infrastructure for the global AI economy. We are building a full-stack AI cloud platform that supports developers and enterprises from data and model training through to production deployment, without the cost and complexity of building large in-house AI/ML infrastructure. Built by engineers, for engineers. From large-scale GPU orchestration to inference optimization, we own the hard problems across compute, storage, networking and applied AI. Listed on Nasdaq (NBIS) and headquartered in Amsterdam, we have a global footprint with R&D hubs across Europe, the UK, North America and Israel. Our team of 1,500+ includes hundreds of engineers with deep expertise across hardware, software and AI R&D. About the Product Token Factory is focused on building a next-generation platform that enables companies to seamlessly integrate AI into their products and workflows. Our vision is to create a powerful, open, and scalable alternative for deploying and managing AI systems—making advanced AI infrastructure more accessible to both fast-growing startups and large enterprises. We work with a wide range of customers, from AI-first companies to established technology organisations, helping them run AI workloads reliably at scale. Our goal is to become a leading platform for high-performance AI inference, delivering predictable latency, strong reliability, and the ability to scale to meet demanding production needs. Customer feedback plays a central role in how we build—our development process is highly iterative and closely aligned with real-world use cases. The Models Team is responsible for onboarding state-of-the-art (SOTA) open-source models into Nebius TokenFactory, including models such as DeepSeek V4 Pro, GLM 5.1, Kimi K2.6, and Minimax M2.7. A major focus of the team is serving large-scale AI models efficiently and reliably in production. To achieve this, we work on advanced inference and systems optimization techniques, including: Cache-aware routing NUMA-aware deployments KV-cache offloading Disaggregated serving architectures Autoscaling with high-speed model loading over InfiniBand / RoCE The team maintains and extends forks of leading inference frameworks such as vLLM and TRT-LLM. We have deep expertise in production-scale model serving and regularly support the Solutions Architects team on the most demanding customer PoCs. To operate efficiently at scale, we invest heavily in tooling and automation.