Responsibilities

Team Introduction: E-commerce is a new and fast growing business that aims at connecting all customers to excellent sellers and quality products, through E-commerce live-streaming, E-commerce short videos, and commodity recommendation. Our E-commerce Recommendation Infra team is responsible for building up and optimizing the infrastructure for such recommendation systems, so as to provide the best experience for our users. We work closely with applied machine learning engineers and build scalable systems to support all kinds of innovative algorithms and techniques. Project introduction With the rapid development of LLM technologies, traditional deep learning algorithms and system architectures for recommendation are facing both transformational challenges and opportunities: - (1) Lifelong Behavior Modeling and Model Scale-up - (2) End-to-End Generative Recommendation We are looking for talented individuals to join our team. As a graduate, you will get opportunities to pursue bold ideas, tackle complex challenges, and unlock limitless growth. Successful candidates must be able to commit to an onboarding date by the end of the year. Please state your availability and graduation date clearly in your resume. Candidates can apply to a maximum of two positions and will be considered for jobs in the order you apply. The application limit is applicable to our Company and its affiliates' jobs globally. Applications will be reviewed on a rolling basis - we encourage you to apply early. Responsibilities 1. Deeply integrate LLM and recommendation technologies; develop, deploy, and optimize LLM training and inference techniques for recommendation foundation models; and solve large-model engineering challenges in recommendation scenarios. 2. Provide highly automated, agentic, and performance-oriented model optimization solutions for frameworks such as PyTorch. 3. Build, design, and implement next-generation model infra architectures based on open-source components like Vllm/SGlang/Megatron/VeRL, enabling recommendation foundation models to be deployed in production. 4. Co-design with algorithm teams to research, innovate, and productionize end-to-end generative recommendation technologies. 5. Research Areas: (1) LLM-native recommendation models and training/inference architectures - Re-architect recommendation models and the complete training/inference technology stack through secondary development based on PyTorch and LLM components. - Ultra-long-context modeling of users' lifelong behavioral histories. - KV cache sharing across multiple user browsing sessions. - Multimodal recommendation foundation models and downstream fine-tuning tasks. (2) Automated, agentic, and extreme-performance model optimization - Throughput optimization for high-performance GPU kernels and multi-GPU parallelism. - Extreme performance optimization tailored to the specialized structures of recommendation models. - Agentic paradigms for performance optimization. - Exploration of new AI infrastructure paradigms enabled by AI-powered productivity. (3) Next-generation recommendation system architectures built on foundation models - Shared modeling and shared computation across retrieval, pre-ranking, ranking, and re-ranking stages. - Prefill/decode disaggregation and computation graph partitioning tailored to recommendation workloads. - LLM architectures that support the sparse characteristics of recommendation scenarios. (4) Innovation in end-to-end generative recommendation - Alignment between multimodal LLM foundations and recommendation tasks. - Implementation, deployment, and optimization of reinforcement learning techniques in recommendation scenarios. - Data and sample construction for generative recommendation. - Exploration of new recommendation paradigms and technical approaches.

Qualifications

Minimum Qualifications: - Individuals who are completing or have recently completed a Bachelor's or Master's degree in Artificial Intelligence, Software Development, Computer Science, Computer Engineering or a related discipline. - Solid programming skills in C++/CUDA/Triton/Python. - Familiarity with GPU architecture and distributed training is highly desirable. - Experience programming in at least one of the following programming languages: C, C++, Java or Golang. - Effective communication skills and a sense of ownership and drive. Preferred Qualifications: - Strong programming and algorithmic foundations, with proficiency in languages such as C/C++, Python. Experience with CUDA development and familiarity with TensorRT, Triton, or CUTLASS is preferred. - Familiarity with research and technical developments in LLM inference acceleration, including but not limited to model quantization, graph compilation, operator optimization, KV cache optimization, and prefill/decode disaggregation. - Hands-on knowledge of LLM training and inference technologies, with practical development and production experience using systems such as Megatron-LM, DeepSpeed, vLLM, SGLang, or TensorRT-LLM. - Extensive experience and broad technical perspective in foundation model engineering, close awareness of open-source developments, and strong independent troubleshooting capabilities. - Experience in recommendation, advertising, or search model development and optimization, or contributions to open-source foundation model communities, is a plus.

Job Information