Participate in building domain LLM model capabilities for e-commerce supply chain and logistics, including continued pre-training / CPT, SFT, preference optimization, reinforcement learning such as GRPO / PPO, reward or judge model design, model compression, inference cost and latency optimization, and landing scenarios such as address correction, trajectory prediction, logistics cost analysis, customer service semantic understanding, and root-cause analysis. Support agent architecture, engineering, evaluation, and evolution work, including runtime orchestration, memory and state management, model / tool routing, permission-safe execution, observability, benchmark and Golden Set evaluation, bad case attribution, regression testing, online feedback loops, and continuous improvement of context, skills, workflows, and model behavior.