Job Responsibilities
1. Frontier Tech Exploration: Conduct deep optimization of LLM/VLM Agent architectures. Lead research in cutting-edge topics including Multimodal GUI Agents, Web Agents, and Long-term Memory mechanisms.
2. Agent Evolution (RL): Explore the application of Agentic RL in both real-world and simulated environments. Solve critical challenges such as policy drift in multi-turn long-horizon interactions, self-evolving mechanisms (Self-reflection/Discovery), and large-scale exploration efficiency.
3. Foundation Model Breakthroughs: Participate in the iteration of next-generation Agentic Foundation Models (e.g., MAI-UI 2.0). Significantly enhance model success rates in fine-grained UI perception, complex logical reasoning, and long-horizon planning.
4. Business Value Closure: Track SOTA (State-of-the-Art) developments to define and implement high-value scenarios for To-C personal assistants or To-B automation workflows, building AI-native applications with strong technical moats.
Job Qualifications
1. Academic Background: Master's or Ph.D. in Computer Science, AI, Mathematics, or related fields. Solid track record of research or high-level publications in LLM/VLM, Planning & Reasoning, or Reinforcement Learning.
2. Algorithm Depth & Breadth: Familiar with mainstream Large Model architectures (Transformers, MoE). Hands-on experience with Chain-of-Thought (CoT), Agent frameworks (e.g., LangGraph, AutoGPT), Tool-use augmentation, and RL algorithms (PPO, DPO, etc.) is highly preferred.
3. Hardcore Engineering: Strong systems thinking with the ability to handle convergence instability and hallucination suppression in large-scale training. Familiarity with inference acceleration, high-concurrency environment simulation, and Sandbox virtualization for Agent infrastructure.
4. Language & Collaboration: Proficiency in English and Mandarin (Chinese) is required. This role involves daily technical collaboration, code reviews, and architectural discussions with our R&D teams in China.
5. Geek Spirit: Exceptional technical curiosity and self-drive. Proven ability to define ambiguous problems from 0 to 1 and maintain leadership in rapidly evolving industry benchmarks.
6. Minimum 1 year of experience.
Bonus Points:
1. Academic/Community Influence: First-author publications in top-tier conferences (NeurIPS, ICLR, ICML, CVPR, etc.) or lead contributor to renowned open-source projects (GitHub 1k+ Stars).
2. Competition Honors: ACM/ICPC regional gold medals or above, or top rankings in LLM/Agent-related challenges (e.g., GAIA, OSWorld, SWE-bench).
3. System-level Expertise: Experience leading the full-cycle training of models with 3B+ parameters. Deep customization experience with distributed frameworks like Megatron-LM or DeepSpeed.
4. Cross-stack Capabilities: Background in low-level system development, with familiarity in browser engines (Chromium/WebKit), Linux kernel, or containerization.