AI Infrastructure Engineer

NIO Inc. is a pioneer and a leading company in the global smart electric vehicle market.

Founded in November 2014, NIO aspires to shape a sustainable and brighter future with the mission of “Blue Sky Coming”.

NIO envisions itself as a user enterprise where innovative technology meets experience excellence.

NIO designs, develops, manufactures and sells smart electric vehicles, driving innovations in next-generation core technologies.

NIO distinguishes itself through continuous technological breakthroughs and innovations, exceptional products and services, and a community for shared growth.

NIO provides premium smart electric vehicles under the NIO brand, premium smart electric vehicles for families through the ONVO brand, and small smart high-end electric cars with the FIREFLY brand.

• Design and implement high-performance, scalable inference systems for LLMs and VLMs across cloud, edge, and edge-cloud hybrid platforms.
• Develop and optimize custom kernels and operators for specific hardware accelerators (GPU, NPU, DSP, etc.), improving throughput, latency, and memory efficiency.
• Integrate advanced optimization techniques such as KV-cache management, tensor/model parallelism, quantization, and memory-efficient execution into production inference systems.
• Partner with system and hardware teams to ensure tight hardware-software integration and optimal performance across diverse compute environments.
• Translate architectural requirements into robust, maintainable, production-ready software that meets performance, safety, and reliability standards.
• Define and drive the evolution roadmap for LLM/VLM inference in the AIOS stack, ensuring scalability and adaptability to new workloads.
• Stay ahead of industry trends and competitor solutions, applying best practices from both AI and large-scale systems engineering.
• 5+ years of hands-on software development experience in building and optimizing AI inference systems at scale.
• Direct experience in LLM/VLM model internals, including Transformer-based architectures, inference bottlenecks, and optimization techniques.
• Strong expertise in performance engineering: kernel development, parallelism strategies, memory optimization, and distributed inference systems.
• Proficiency with GPU/NPU programming (CUDA, or vendor-specific SDKs), compiler toolchains, and deep learning frameworks (PyTorch, or TensorFlow).