大模型算法工程师（J105959）
Core
Pretrain and fine-tune multimodal large models for map applications, research reward models and RLHF/RLAIF to enhance domain reasoning, and build map production agents with task planning and multi-agent collaboration.
Role type
Senior IC large model algorithm engineer (multimodal/agents/maps)
Builds
Multimodal large models, map production agents, and reinforcement learning reward systems for autonomous decision-making in complex map scenarios.
Domain
Autonomous driving, mapping, and embodied intelligence
Deliverable
production ML models
Required skills
Multimodal large model algorithms (CLIP, Diffusion, VLM), Reinforcement learning (PPO, DPO, GRPO), Agent development (Function Calling, MCP, Tool Use, Multi-Agent, Planning, Reasoning, RAG), Open-source model training (Qwen, LLaMA, DeepSeek), Complex system design for agent architecture
Preferred skills
End-to-end large model pretraining experience in autonomous driving, High-level paper publications, Experience with mainstream agent frameworks or open-source projects, GitHub or competition project experience
Technologies
Qwen, LLaMA, DeepSeek, CLIP, Diffusion, PPO, DPO, GRPO, MoE, MCP, RAG
Responsibilities
Pretrain and fine-tune multimodal large models for map scenarios, Research vertical domain reward rules and RLHF/RLAIF techniques, Build map production agents with task planning and multi-agent collaboration, Research agent memory, world models, and agent workflow technologies, Explore frontier model architectures and efficient training/inference techniques
Seniority
Senior, hands-on IC
