Staff Software Engineer
On-siteBeijing, Beijing, China
Job Summary
Fine-tune large language models using supervised fine-tuning, reinforcement learning from human feedback, and domain-specific adaptation. Design high-efficiency prompt templates, optimize function calling capabilities, and implement advanced reasoning technologies like Chain of Thought and long context management. Build a comprehensive evaluation system to test model hallucination, stability, and tool call accuracy. Develop the core Agent runtime system, including execution engines, state machines, and memory modules, while optimizing scheduling logic and multi-Agent collaboration mechanisms. Connect and customize open-source frameworks such as LangChain and AutoGPT to support complex business scenarios. Participate in agile processes, planning, estimation, and retros, with on-call support as needed.
Required Qualifications
- MS or PhD degree or above in Computer Science, Artificial Intelligence, Mathematics, Statistics and other related majors
- More than 3 years of relevant working experience in large language model algorithm research and development
- Familiar with the training, fine-tuning and inference process of mainstream open-source large models (such as LLaMA, Qwen, ChatGLM series, etc.)
- Proficient in deep learning frameworks such as PyTorch, TensorFlow
- Familiar with model fine-tuning tools such as PEFT, LoRA
- Hands-on experience in large model parameter-efficient fine-tuning
- In-depth understanding of Prompt Engineering, Function Calling, Chain of Thought and other LLM related technologies
- Practical project experience in model reasoning optimization and long context processing
- Proficient in Python programming
- Good data structure and algorithm foundation
- Strong code implementation and problem-solving abilities
- Strong learning ability
- Innovative thinking
- Ability to independently tackle key technical problems
- Good communication and collaboration skills
- Ability to efficiently cooperate with cross-team members
- Ability to promote project progress
Desired Qualifications
- Experience in building LLM evaluation systems
- Conducting model hallucination, stability and tool call accuracy testing
- Nice to Have CUDA/CuFFT/CuBLAS
- OpenMP, SIMD vectorization
- Experience with git, jenkens, devops, etc.
- Publications/competitions or open-source contributions
Hiring someone like this?
Get your role in front of qualified candidates on Sorce.