I am Kai Wang (王凯), currently a Researcher at TeleAI, focusing on large-scale foundation model security and trustworthy AI. My research centers on building systematic defense mechanisms across the data, model, and applications of modern LLMs. My primary research interests include: Data Provenance, Attribution, and Backtracking; Self-Repairing Large Models & Robustness Enhancement; Personalized & Multi-value Defensive Alignment; Cross-Modal Safety Alignment; Safe Decoding & Risk-Aware Generation.

Before joining TeleAI, I served as a Senior Algorithm Expert at ByteDance and Kuaishou, where I led mission-critical projects such as Kuaishou KuaYi LLM Security and TikTok Shop Logistics Compliance. These systems handled hundreds of millions of requests per day, requiring millisecond-level latency, high-precision risk control, and resilience against continuously evolving adversarial behaviors. Through systematic model–data–rule co-design, the solutions I built achieved industry-leading safety performance and were deployed at global scale, enabling large-model technologies to operate reliably within high-stakes, real-world industrial ecosystems.

During my early career, I received multiple prestigious talent awards, including Huawei TopMinds (华为天才少年), Kuaishou “KuaiStar-X” (快手快Star), Tencent “Qingyun” (腾讯青云), and Meituan “Beidou” (美团北斗), and ACM Second Prize Award.

🔥 News

  • 2025.11:  🎉🎉 I joined TeleAI as a Researcher.

📝 Selected Publications

🎖 Honors and Awards

  • National Scholarship

📖 Educations

  • 2018.09 - 2021.06, Master of Computer Science and Technology, Sun Yat-sen University
  • 2014.09 - 2018.06, Bachelor of Network Engineering, Sun Yat-sen University

💬 Academic Service

  • ACL 2021, Reviewer
  • EMNLP 2020, Reviewer
  • AAAI 2019, Reviewer
  • NLPCC 2018, Reviewer

💻 Internships

  • 2020.04 - 2020.09, Tencent AI LAB (腾讯AI LAB), China.
  • 2019.07 - 2019.12, Alibaba DAMO Academy (阿里巴巴达摩院), China.
  • 2017.03 - 2017.07, Baidu (百度), China.