AI 基础设施工程师

We are looking for AI Infrastructure Engineers to build the data infrastructure, AI development platform, MLOps/LLMOps toolchains, model serving systems, and agent infrastructure; The Engineers in this role will work closely with applied AI engineers, scientists and domain experts to understand concrete research bottlenecks and translate them into reliable, reusable and scalable AI engineering solutions.This role spans three complementary tracks. You are not expected to cover all of them. Each engineer will take primary ownership of ONE to TWO tracks based on their experience, strengths, and team needs.我们正在招聘 AI 基础设施工程师，负责建设现代 AI 工程及研发团队所需的数据基础设施、AI 开发平台、MLOps/LLMOps 工具链、模型基础设施和智能体基础设施。该岗位的工程师将与AI应用工程师、科学家、领域专家紧密合作，理解具体科研瓶颈，并将其转化为可靠、可复用、可扩展的AI工程解决方案。该岗位涵盖三个相互协同的方向。我们不期待任何一位候选人同时覆盖所有方向。*每位工程师将根据自身经验、优势和团队需求，主要负责其中一到两个方向。岗位职责：Track 1: Data Engineering方向一：数据工程- Design and build scalable data pipelines for AI model development, training, evaluation, and production applications. 设计并建设可扩展的数据 Pipeline，支持 AI 模型开发、训练、评估和生产应用。- Build ingestion, transformation, validation, versioning, and lineage capabilities for structured and unstructured data. 建设结构化与非结构化数据的采集、转换、校验、版本管理和数据血缘能力。- Develop reliable workflows for processing text, images, documents, scientific data, time-series data, and other domain-specific datasets. 开发可靠的数据处理流程，支持文本、图像、文档、科学数据、时序数据及其他领域数据。- Build data infrastructure for LLM and RAG applications, including vector stores, knowledge bases, chunking and embedding pipelines, and retrieval quality evaluation. 建设面向大模型与 RAG 应用的数据基础设施，包括向量存储、知识库、切分与向量化 Pipeline，以及检索质量评估。- Improve data quality, observability, reproducibility, and accessibility across AI development workflows. 提升 AI 开发流程中的数据质量、可观测性、可复现性和可访问性。- Work with algorithm and domain teams to translate model development requirements into reusable data infrastructure. 与算法团队和业务领域团队合作，将模型开发需求转化为可复用的数据基础设施。Track 2: AI Platform, MLOps/LLMOps and Model Infrastructure方向二：AI 平台、MLOps/LLMOps 与模型基础设施- Design and build AI development platforms that support the end-to-end model lifecycle, including data preparation, experimentation, training, evaluation, deployment, inference, monitoring, and iteration. 设计并建设支持模型全生命周期的 AI 开发平台，包括数据准备、实验、训练、评估、部署、推理、监控和持续迭代。- Build standardized MLOps and LLMOps pipelines, tools, and workflows for model development, lifecycle management, and production delivery. 建设标准化的 MLOps 和 LLMOps Pipeline、工具及工作流，支持模型开发、全生命周期管理和生产交付。- Provide reusable platform capabilities for experiment tracking, dataset and model versioning, model evaluation, model registry, release management, and reproducibility. 提供可复用的平台能力，支持实验追踪、数据集与模型版本管理、模型评估、模型注册、发布管理和结果复现。- Build scalable training, fine-tuning, evaluation, and inference workflows across local, cloud, Kubernetes, GPU cluster, and hybrid environments. 建设适用于本地、云端、Kubernetes、GPU 集群及混合环境的可扩展训练、微调、评估和推理工作流。- Develop standardized model deployment and serving capabilities, including model packaging, rollout, request routing, load balancing, autoscaling, monitoring, and failure recovery. 开发标准化的模型部署与服务能力，包括模型封装、发布、请求路由、负载均衡、弹性扩缩、监控和故障恢复。- Integrate and operate mainstream model training and serving frameworks such as PyTorch, DeepSpeed, FSDP, vLLM, SGLang, TensorRT-LLM, Triton Inference Server, or equivalent technologies. 集成并运行主流模型训练和服务框架，例如 PyTorch、DeepSpeed、FSDP、vLLM、SGLang、TensorRT-LLM、Triton Inference Server 或同类技术。- Optimize model training and inference systems for performance, scalability, GPU utilization, memory efficiency, reliability, and cost. 围绕性能、可扩展性、GPU 利用率、显存效率、可靠性和成本，优化模型训练与推理系统。- Improve AI developer experience by providing standardized development environments, SDKs, APIs, templates, CI/CD pipelines, and self-service platform capabilities. 通过标准化开发环境、SDK、API、模板、CI/CD Pipeline 和自助式平台能力，提升 AI 开发者体验。- Build observability, resource scheduling, capacity management, cost monitoring, security, and governance capabilities for AI workloads. 建设 AI 工作负载的可观测性、资源调度、容量管理、成本监控、安全及治理能力。- Collaborate with research, data, and application teams to translate model development requirements into reusable platform and infrastructure capabilities. 与研发、数据及应用团队协作，将模型开发需求转化为可复用的平台与基础设施能力。Track 3: Agent Infrastructure方向三：智能体基础设施- Design and build infrastructure for developing, deploying, operating, and evaluating AI agents. 设计并建设支持 AI 智能体开发、部署、运行和评估的基础设施。- Build reusable agent runtime capabilities, including model access, tool invocation, workflow orchestration, state management, memory, and execution control. 建设可复用的 Agent Runtime 能力，包括模型接入、工具调用、工作流编排、状态管理、记忆和执行控制。- Develop infrastructure for agent identity, permissions, sandboxing, secrets management, observability, tracing, and auditability. 建设智能体身份、权限、沙箱、密钥管理、可观测性、链路追踪和审计能力。- Build agent evaluation and testing systems covering task completion, tool-use correctness, reliability, safety, latency, and cost. 建设智能体评估与测试系统，覆盖任务完成度、工具调用正确性、可靠性、安全性、延迟和成本。- Provide reusable SDKs, APIs, templates, skills, connectors, and deployment workflows for agent development teams. 为智能体开发团队提供可复用的