首席机器学习工程师
Principal Machine Learning Engineer
关于公司
我们的客户是一家由东南亚领先科技公司支持的隐秘人工智能初创公司，目前正在组建其全球创始团队。
该公司正在开发一个以人工智能为核心的通信平台，旨在通过将人工智能直接融入对话中，简化日常任务。用户无需在多个应用程序之间切换，即可在一个智能助手内完成计划、组织、比较、研究和完成任务。
面向仍依赖传统生产力工具的数十亿用户市场，该平台专注于提供可靠的AI工作流程、持久的上下文、多步骤推理和无缝的任务执行。其使命是创造一个能显著提升生产力，同时让日常办公更简单、更直观的人工智能助手。
关于职位
我们的客户正在寻找一位首席机器学习工程师，负责构建和部署为AI平台提供支持的生产级机器学习系统。该职位的重点是通过开发稳健的训练管道、推理系统、评估框架和部署基础设施，将研究成果转化为可扩展的解决方案。
与研究和应用工程团队紧密合作，该职位将在交付可靠、高性能的ML系统方面发挥关键作用，这些系统能够在实际生产约束下有效运行。
主要职责
- 构建并负责端到端的机器学习管道，涵盖数据处理、模型训练、评估、推理和部署。
- 使用LoRA、QLoRA、监督微调（SFT）、直接偏好优化（DPO）和模型蒸馏等现代技术对模型进行微调和适应。
- 设计和运营可扩展的推理系统，同时平衡延迟、成本和可靠性。
- 开发和维护用于合成和真实世界训练数据集的数据管道。
- 与研究团队合作，构建评估框架，评估模型性能、鲁棒性、安全性和偏见。
- 通过GPU优化、内存效率、延迟降低和扩展策略优化生产部署。
- 与应用工程团队合作，将机器学习系统集成到后端、移动端和桌面应用中。
- 通过快速迭代和现实世界性能监控持续改进ML系统，同时平衡生产约束，如延迟、成本、可靠性和安全性。
要求
- 强烈的…
查看英文原文
About the Company
Our client is a stealth AI startup backed by one of Southeast Asia's leading technology companies and is currently building its global founding team.
The company is developing an AI-native communication platform designed to simplify everyday tasks by integrating AI directly into conversations. Instead of switching between multiple applications, users can plan, organize, compare, research, and complete tasks within a single intelligent assistant.
Serving a market of billions of users still relying on traditional productivity tools, the platform focuses on delivering reliable AI workflows, persistent context, multi-step reasoning, and seamless task execution. The mission is to create an AI assistant that significantly improves productivity while making everyday work simpler and more intuitive.
About the Role
Our client is seeking a Principal Machine Learning Engineer to build and deploy production-grade machine learning systems that power its AI platform. This role focuses on translating research into scalable solutions by developing robust training pipelines, inference systems, evaluation frameworks, and deployment infrastructure.
Working closely with research and application engineering teams, this position will play a key role in delivering reliable, high-performance ML systems that operate effectively under real-world production constraints.
Key Responsibilities
- Build and own end-to-end machine learning pipelines covering data processing, model training, evaluation, inference, and deployment.
- Fine-tune and adapt models using modern techniques such as LoRA, QLoRA, Supervised Fine-Tuning (SFT), Direct Preference Optimization (DPO), and model distillation.
- Design and operate scalable inference systems while balancing latency, cost, and reliability.
- Develop and maintain data pipelines for both synthetic and real-world training datasets.
- Build evaluation frameworks to assess model performance, robustness, safety, and bias in collaboration with research teams.
- Optimize production deployments through GPU optimization, memory efficiency, latency reduction, and scaling strategies.
- Collaborate with application engineering teams to integrate machine learning systems into backend, mobile, and desktop applications.
- Continuously improve ML systems through rapid iteration and real-world performance monitoring while balancing production constraints such as latency, cost, reliability, and safety.
Requirements
- Strong background in deep learning and transformer-based architectures.
- Hands-on experience training, fine-tuning, or deploying large-scale machine learning models in production.
- Proficiency with modern machine learning frameworks such as PyTorch or JAX.
- Experience with distributed training and inference frameworks, including technologies such as DeepSpeed, FSDP, Megatron, ZeRO, or Ray.
- Strong software engineering skills with experience building robust, maintainable, production-grade systems.
- Experience optimizing GPU workloads, including memory efficiency, quantization, and mixed precision.
- Ability to independently own end-to-end machine learning systems in fast-moving environments.
- Strong problem-solving skills with a focus on rapid iteration and continuous improvement.
Preferred Qualifications
Experience with one or more of the following is preferred:
- LLM inference frameworks such as vLLM, TensorRT-LLM, or FasterTransformer
- Open-source contributions to machine learning or systems libraries
- Scientific computing, compiler technologies, or GPU kernel development
- Reinforcement Learning from Human Feedback (RLHF) pipelines, including PPO, DPO, or ORPO
- Training or deploying multimodal or diffusion models
- Large-scale data processing frameworks such as Apache Arrow, Spark, or Ray
Originally posted on Himalayas