About the job
We are building next-generation Trust & Safety systems powered by large language models and multimodal foundation models to protect TikTok, Lemon8, and short-form video platforms. Our team focuses on multimodal reasoning, content understanding, and large-scale risk detection across text, image, audio, and video. We actively explore and productionize cutting-edge technologies such as LLM post-training, multimodal models, and AI Agents to build intelligent, scalable, and adaptive content moderation systems.
Responsibilities
- Develop content safety algorithms for TikTok, Lemon8, and short-form drama platforms, covering multimodal understanding and risk detection across video, text, image, and audio
- Leverage machine learning, deep learning, and large model technologies to improve content moderation pipelines, enhance operational efficiency, and ensure model robustness and production stability
- Own the training and application of LLMs and MLLMs for content safety use cases, including post-training techniques such as SFT, RLHF, DPO, and safety alignment
- Drive the exploration and productionization of emerging capabilities such as AI Agents, multimodal reasoning, and video understanding to build next-generation intelligent moderation systems
Qualifications
Minimum
- Bachelor degree or above in Computer Science, Software Engineering, Artificial Intelligence, Mathematics, or related fields
- Strong understanding of machine learning and deep learning, with familiarity with Transformer architectures, LLMs, and MLLMs
- Hands-on experience in Python, PyTorch, distributed training, and large-scale data processing, with solid algorithmic and engineering capabilities
- Proven ability to solve complex problems and collaborate effectively in cross-functional teams, with a track record of driving projects to high-quality execution
Preferred
- Hands-on experience with LLM post-training, multimodal understanding, AI Agents, or Trust & Safety systems is desirable
- Publications in academic conferences such as NeurIPS、ICML、ICLR、CVPR、ECCV、ICCV、ACL、EMNLP、NAACL、Interspeech、ICASSP, and others are a strong plus