后训练算法专家（J104317）
Core
Develop and optimize algorithms for large language models (LLMs) in document processing, search, content creation, research analysis, and complex office scenarios, focusing on instruction understanding, complex reasoning, tool use, and long-context capabilities.
Role type
Senior IC post-training algorithm engineer (LLM)
Builds
Production-ready LLMs and inference strategies for enterprise productivity and search applications
Domain
Artificial Intelligence / Large Language Models / NLP
Deliverable
production ML models
Required skills
Transformer architecture, SFT, DPO, RLHF, GRPO, model distillation, Python, PyTorch, data cleaning, synthetic data generation, LLM-as-Judge evaluation, model routing, inference optimization
Preferred skills
Tool-use/Agentic models, long-context training, preference data construction, hard example mining, AI search, document understanding, open-source model expertise (Qwen, Llama), training frameworks (DeepSpeed, vLLM)
Responsibilities
Build post-training data systems including online hard example mining and trajectory analysis; Implement alignment schemes to improve model stability and generalization; Construct evaluation systems with offline datasets and automated assessors; Analyze user task failures to create iterative training loops; Optimize model selection and inference strategies for cost-latency balance; Collaborate with product and engineering teams for stable online deployment
Seniority
Senior, hands-on IC
