Xiaohongshu Hiring! Full Time Agent - VLM Algorithm Engineer – International Trust - Safety in - Ricebowl

Job Responsibilities:

1. Leverage multimodal AI—VLMs and Agent systems—to fuse signals across text, images, video, and audio from user-generated content. Drive international content risk detection and enforcement spanning posts, comments, accounts, commerce, and livestreams, continuously improving proactive risk detection and recall while balancing risk control, user experience, and operational efficiency.
2. Build and continuously evolve rednote's content representation and classification systems. Apply these capabilities to core governance workflows—enforcement decisions, penalties, search, recommendation, and traffic control—to establish durable, scalable foundations for content quality and safety across rednote's global ecosystem.
3. Own end-to-end R&D of Agent systems for rednote's core scenarios including governance, content understanding, and editing—covering ReAct / Tool Use / Multi-Agent architectures, intent understanding, and task planning. Drive Agent performance optimization across tool design, context management, orchestration strategies, and post-training with Agentic RL to deliver industry-leading products.
4. Own key stages of the LLM/VLM training pipeline—pre-training, SFT, and RL—to advance foundational multimodal understanding, editing, and generation capabilities. Close the loop from AI research to measurable business impact.
5. Continuously track the latest developments in the AI Agent field, introduce and validate new technologies, synthesize technical insights and best practices into reusable frameworks, and drive the adoption of cutting-edge research in rednote's business scenarios to maintain the team's technological leadership. Contribute to publications at top international conferences such as CVPR, NeurIPS, ICLR, and ACL.

Job Requirements

1. Master's degree or above in Computer Science, Artificial Intelligence, Machine Learning, Mathematics, or related fields
2. Deep understanding of large language models and Agent systems, including deployment and inference paradigms, prompt engineering, RAG, and evaluation frameworks. Familiarity with mainstream Agent architectures—Multi-Agent, Context Engineering/Management, ReAct/PlanAct/CodeAct—and interaction protocols such as MCP, A2A, and Function Calling
3. Solid grounding in Agentic RL and related infrastructure (e.g., CodeRL, reward modeling). Hands-on experience with SFT and RL training pipelines, with the ability to drive optimization for real-world use cases
4. Strong academic vision and pioneering spirit, with a genuine curiosity for exploring new technologies. Publications at top conferences such as CVPR, ICCV, NeurIPS, ICLR, ACL, EMNLP, or KDD are a plus

.