The Tencent Overseas IT team is seeking a Software Engineer II to join their mission of empowering global growth through future-ready IT platforms and services. This role focuses on the design, development, and delivery of GPU and AI infrastructure platforms. Responsibilities include building core capabilities for GPU resource management, workload scheduling, distributed training, and inference. You will develop and improve engineering infrastructure for mainstream AI frameworks and runtimes specifically for LLM workloads, while optimizing GPU cluster performance, stability, and resource utilization. You will also collaborate with algorithm and business teams to solve complex AI workload engineering challenges. Required Qualifications: Strong software engineering skills in Python, Go, or C++; solid understanding of Linux, Docker, and Kubernetes; familiarity with GPU technologies including CUDA, NCCL, and RDMA; and experience with distributed systems or distributed computing. You should also have experience with AI coding, AI agents, or AI-assisted software development. Preferred Qualifications: Experience in LLM training, inference, or GPU cluster infrastructure is considered a plus. Tencent is an equal opportunity employer committed to fostering a diverse and supportive work environment.
For full details about this role, click “Apply on company site” to view the employer’s original posting.
A focused preparation guide, built around this job’s responsibilities and requirements.
✦ AI-generated guide
Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.
01
Understand the role
What matters most
GPU Infrastructure Development
The core of this role is designing and delivering platforms for GPU resource management and workload scheduling. You must be prepared to demonstrate expertise in building scalable systems that handle distributed training and inference, specifically optimizing for performance, stability, and resource utilization in large-scale environments.
AI Framework and Runtime Optimization
You will be responsible for improving engineering infrastructure for mainstream AI frameworks and runtimes tailored for LLM workloads. This requires a deep understanding of how these frameworks interact with underlying hardware and the ability to tune them for maximum efficiency.
Cross-Functional Collaboration
Success in this role depends on working closely with algorithm and business teams to solve complex engineering challenges. You must be able to translate high-level AI workload requirements into robust, efficient infrastructure solutions while maintaining clear communication with non-engineering stakeholders.
Likely challenges to think through
Balancing Performance with Resource Utilization
Optimizing GPU clusters often involves a trade-off between raw performance and cost-effective resource utilization. Prepare to discuss strategies for managing these competing priorities, such as implementing efficient scheduling algorithms or optimizing data movement via RDMA and NCCL.
Adapting to Rapidly Evolving AI Workloads
The field of LLM training and inference is moving quickly, which may lead to shifting requirements for your infrastructure. Prepare by demonstrating a flexible approach to system design that can accommodate new AI frameworks or changing hardware capabilities without requiring a complete architectural overhaul.
03
Practice your answers
Technical practice
GPU Resource Management and Scheduling
What to coverExplain how you would design a scheduler to manage GPU resources for heterogeneous AI workloads. Discuss the trade-offs between static partitioning and dynamic oversubscription in a Kubernetes-based environment.
Optimizing Distributed Training with NCCL
What to coverDescribe the role of NCCL in distributed training and how you would troubleshoot performance bottlenecks in a multi-node GPU cluster. Focus on how RDMA integration impacts communication latency.
LLM Inference Infrastructure
What to coverWhat are the primary challenges in deploying LLMs for high-throughput inference? Discuss strategies for model parallelism, KV cache management, and optimizing GPU memory utilization.
Kubernetes for AI Workloads
What to coverHow do you handle GPU device plugins and custom resource definitions (CRDs) in Kubernetes to ensure efficient workload scheduling? Suggest how to monitor and enforce resource quotas for AI training jobs.
AI-Assisted Development Tools
What to coverDescribe your experience integrating AI agents or coding assistants into a CI/CD pipeline. How do you evaluate the reliability and security of code generated by AI tools in a production environment?
Linux Kernel and Performance Tuning
What to coverWhich Linux kernel parameters or system-level configurations are critical for high-performance GPU computing? Explain how you would identify and resolve I/O or memory contention issues in a distributed system.
Distributed Systems Consistency
What to coverIn a distributed training setup, how do you handle node failures or network partitions without losing significant progress? Discuss the trade-offs between checkpointing frequency and training throughput.
Programming for AI Infrastructure
What to coverCompare the use of Python, Go, and C++ in the context of building AI infrastructure platforms. When would you choose one over the others for developing low-level runtime components versus high-level orchestration tools?
Behavioural practice
Cross-functional Collaboration
What to coverDescribe a time you worked with algorithm or data science teams to solve a technical infrastructure challenge. How did you translate their requirements into actionable engineering tasks?
Handling Production Stability Issues
What to coverTell me about a time you had to diagnose a performance degradation in a large-scale cluster. What was your process for isolating the root cause and implementing a long-term fix?
Adapting to New AI Technologies
What to coverThe AI field evolves rapidly. Provide an example of how you recently learned and applied a new framework or technology to improve your infrastructure work.
Prioritizing Technical Debt
What to coverHow do you balance the need for rapid delivery of new AI infrastructure features with the necessity of maintaining system stability and reducing technical debt?
02
Sharpen your knowledge
Topics to prepare
Distributed GPU Training Architectures
Review the principles of data parallelism and model parallelism in the context of LLM training. Exercise: Diagram how NCCL facilitates communication between GPUs in a multi-node cluster and explain how this impacts training latency. This is critical for the role's focus on building distributed training infrastructure.
Kubernetes Scheduling for AI Workloads
Study advanced Kubernetes scheduling concepts, specifically custom schedulers and resource quotas for GPU-intensive tasks. Exercise: Research how to implement a custom scheduler or use existing frameworks to optimize GPU resource utilization for heterogeneous AI workloads. This directly addresses the requirement for GPU resource management.
Skills to revise
CUDA and RDMA Optimization
Practice profiling GPU kernels and understanding low-latency networking. Exercise: Use Nsight Systems to identify bottlenecks in a simple CUDA-based matrix multiplication and research how RDMA bypasses the CPU to accelerate data transfer between nodes. This is essential for optimizing cluster performance.
AI Agent Integration
Develop a small-scale application using an AI agent framework to automate a software engineering task. Exercise: Build a script that uses an LLM API to analyze code quality or generate unit tests, demonstrating familiarity with AI-assisted development workflows mentioned in the job description.
Go or C++ Systems Programming
Focus on high-performance systems programming. Exercise: Write a Go or C++ service that interacts with the Docker API to manage container lifecycles, simulating the infrastructure tools needed for managing AI runtime environments.
04
Your preparation checklist
Architect a GPU resource management strategy
Prepare a technical walkthrough of how you would design a scheduler for GPU-intensive workloads in a Kubernetes environment. Focus on how you handle resource isolation, bin-packing, and minimizing scheduling latency for distributed training jobs.
Demonstrate distributed systems expertise
Be ready to discuss your experience with NCCL and RDMA in the context of scaling LLM training. Prepare to explain how you identify and resolve bottlenecks in inter-node communication and data throughput within a GPU cluster.
Showcase AI-assisted development workflows
Prepare examples of how you have integrated AI agents or AI-assisted coding tools into your development lifecycle to improve infrastructure reliability or automate routine engineering tasks.
Prepare for deep-dive technical discussions on LLM infrastructure
Review your experience with optimizing inference runtimes. Be prepared to explain trade-offs between different quantization techniques or model serving strategies when optimizing for latency and resource utilization.
Formulate strategic questions for the team
Prepare questions regarding the current scale of their GPU clusters, the primary challenges they face with LLM workload stability, and how the engineering team collaborates with algorithm researchers to bridge the gap between model development and production deployment.
Availability last checked: 18 Oct 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy
Share this opportunityKnow someone who would be a great fit?