Responsibilities
1. Explore technological breakthroughs in the next generation of large model recommendation paradigms, be responsible for tackling the technical architecture of LLM4Rec and Agentic Rec, promote their large-scale implementation in live broadcast recommendation scenarios, and drive leap-forward growth in indicators 2. Create a large representation model with strong semantic understanding, information lossless and full-modal capabilities, explore cutting-edge semantic ID technology and make breakthroughs, and achieve accurate representation of content in complex live broadcast and real-time changing scenarios 3. Promote the in-depth application of large models in scenarios such as recommendation and security governance, explore post-business training, reinforcement learning, agent systems and other technical solutions to reshape each business link and drive breakthroughs in governance efficiency and business indicators 4. Lead the technical construction of large-scale base models in live broadcast scenarios, including exploration of high-quality data pipeline construction, pre-training (CPT), reinforcement learning, and full-modality, to create a unified base that supports recommendation and governance 5. Continue to track and explore cutting-edge technologies, closely follow the definition of research directions for live broadcast business scenarios, and accumulate and promote team technology iterations.
Qualifications
1. Master the technology and application experience related to multi-modal understanding, be familiar with the structure and training framework of mainstream large models, and continue to track the latest progress in related fields 2. Have business passion, strong sense of responsibility, be proactive, and have good communication and teamwork skills 3. Have imagination and independent thinking ability, have excellent experimental analysis and problem-solving skills, be able to propose ideas and put them into practice for verification 4. Applicants who have led projects or published high-level papers (NeurIPS, CVPR, ICCV, ICML, etc.) in the direction of multi-modality or large models are preferred 5. Applicants with end-to-end experience in content understanding in recommendation search or security governance scenarios are preferred.