LLM Algorithm Engineer (LLM 算法工程师)

Location: Beijing / Singapore

About the Role

Improve LLM and agent capabilities through post-training research and experimentation. Work across data strategies, training methods and evaluation to diagnose task performance issues and bring research improvements into products.

Responsibilities

• Research and refine LLM post-training methods and training recipes, designing experiments around target tasks to improve performance and generalization.
• Build and iterate on training data, adjusting quality, distribution and usage strategies in response to capability goals and experimental results.
• Design benchmarks and evaluations that cover target capabilities and practical use cases, providing a consistent basis for comparing training approaches.
• Analyze model behavior and recurring failures, formulate improvement hypotheses, and test them through training experiments while documenting results and their conditions.
• Work with product, engineering and model partner teams to incorporate application feedback into data, training and evaluation, and assess results on real tasks.

Requirements

• Solid foundations in machine learning and LLMs, with research or practical experience in large-scale foundation model training or post-training and familiarity with SFT, reinforcement learning or preference optimization.
• Proficiency in mainstream deep learning frameworks, with the ability to set up experiments independently, analyze training results and diagnose problems.
• Understanding of how data and training methods affect model performance, with the ability to plan improvements based on evaluations and practical use cases.
• Active engagement with relevant research and independent research judgment, with the ability to formulate testable hypotheses from open-ended problems and evaluate them rigorously.

Manus excels at various tasks in work and life, getting everything done while you rest at Manus AI.

What we offer:

• Build at the Frontier of AI Agents - Work with a fast-paced team to turn frontier AI breakthroughs into real-world impact.
• Equity & Shared success - Our share incentive plan enables employees to participate in Manus’s long-term growth and share in the value we create together.
• Unlimited Manus Tokens - Enjoy unlimited Manus tokens to experiment, build, and supercharge your productivity.

If you're passionate about cutting-edge technology and making a real impact, we’d love to hear from you!

-----------------------

工作地点：北京 / 新加坡

岗位概述

围绕大语言模型与智能体的能力提升，开展后训练研究与实验。结合数据策略、训练方法和模型评测，定位影响任务表现的问题，并推动研究成果用于实际产品。

岗位职责

• 研究和优化大语言模型后训练方法与训练方案，围绕目标任务设计实验，提升模型表现与泛化能力。
• 构建和迭代训练数据，结合能力目标与实验结果调整数据质量、分布和使用策略。
• 设计基准测试与评测方法，覆盖目标能力和实际使用场景，为不同训练方案提供可比较的依据。
• 分析模型行为和典型失败案例，提出改进假设，通过训练实验验证，并记录结果与适用条件。
• 与产品、工程及模型合作团队协作，将应用反馈纳入数据、训练和评测迭代，跟进实际任务效果。

任职要求

• 具备扎实的机器学习与大语言模型基础，有大规模的基础模型训练或后训练的研究与实践经验，熟悉监督微调、强化学习或偏好优化等方法。
• 熟悉主流深度学习框架，能够独立搭建实验、分析训练结果并定位问题。
• 理解数据、训练方法与模型表现之间的关系，能够结合评测结果和实际场景制定优化策略。
• 持续跟进相关研究，具备独立研究判断，能够从开放问题中提出可检验的假设，并通过严谨实验验证。