Key Responsibilities

• Support the scaling of cutting-edge research into industry-leading next-generation models by providing large-scale training data acquisition, reinforcement learning (RL) environment construction, and extreme training efficiency optimization.

• Build comprehensive and detailed automated evaluation systems for next-generation models to deepen understanding of capability boundaries and guide future research priorities.

• Apply theoretical breakthroughs to real-world product challenges, driving impactful AI applications.

Basic Requirements

• Strong programming skills, proficient in Python and C/C++ under Linux, familiar with PyTorch and mainstream large model training/finetuning frameworks; able to independently implement complex deep learning models and system modules with strong debugging and performance optimization abilities.

• Experience in large-scale data preprocessing, data generation, and data augmentation; understanding of data-driven model iteration workflows.

• Familiarity with large model training pipelines, including distributed training, model parallelism, and training efficiency optimization.

• Excellent problem-solving skills, collaborative mindset, and strong communication skills.

Preferred Qualifications

• Familiarity with high-performance operator frameworks such as CUDA/Triton/Cutlass.

• Experience with distributed RL frameworks such as veRL / OpenRLHF / Ray.

• Knowledge of large-scale RL environment construction for browser / computer use / code sandbox tasks.

• Experience with distributed training frameworks such as Megatron-Core / Deep- Speed, including multi-node training efficiency tuning and optimization of computation‒communication overlap.

• Outstanding achievements in competitive programming (ACM/ICPC, NOI/IOI, Code- forces, TopCoder).

• Contributions to well-known open-source large model projects or winning results in related competitions.