Post-Training Algorithm Researcher/Engineer
Confirmed on the employer's own hiring board on Oct 17, 2026. First seen by Alion on Sep 4, 2026. Moonshot AI scores C on the Alion truth index.
Job Description:
1. Deeply participate in data processing, cleaning, and optimization for CodeLLM; scientifically improve, filter, and balance the data quality used in LLM pre-training and alignment to enhance the model’s code and agent capabilities during the Pretrain and SFT stages.
2. Actively engage in improving Coding/SWE Agent capabilities through SFT (Supervised Fine-Tuning), RL (Reinforcement Learning), and building/scaling tools, Docker environments, and tasks; personally gain know-how on how to effectively train an agentic model.
3. Deeply explore methods to enhance the model’s problem-solving ability, aiming for: 1) Nearly 100% accuracy on simple and medium-level algorithm problems; 2) Maximizing the solving rate of difficult algorithm problems (IOI/ICPC level), pushing the model’s intelligence upper bound.
4. Improve user problem-solving rates and satisfaction on Code/SWE-related questions, making the model genuinely easy and pleasant to use.
Job Requirements:
1. Bachelor’s degree or above, preferably in Computer Science, Artificial Intelligence, Automation, Mathematics, or related fields.
2. Solid programming foundation, deep understanding of data structures and algorithm design; proficient in one or more mainstream programming languages such as Python and C++; skilled in deep learning frameworks and libraries like PyTorch, TensorFlow, and Hugging Face.
3. Rich hands-on experience in large model pre-training and alignment training, or experience in Code/SWE optimization is preferred.
4. Publication in top-tier academic conferences or outstanding achievements in machine learning and AI fields are preferred.
5. Algorithm competition award experience (e.g., ACM, IOI, NOI, Top Coder) is a plus.
6. Strong sense of responsibility, proactive attitude, and good communication and teamwork skills.