Overview In this role you join TikTok's Trust and Safety team to advance multimodal content moderation at scale. You will work on cutting-edge AI systems that protect billions of users, blending RL, agentic decision-making, and multimodal foundation models. You'll tackle challenges spanning multilingual content, adversarial attacks, and scalable tool integration to improve automated moderation. This is a chance to shape the future of responsible AI in a high-impact, fast-growing environment.
Responsibilities- Research and develop MoE-based multimodal safety models for cross-modal understanding and generation
- Design and train RL-driven agentic moderation systems with end-to-end reasoning and tool collaboration
- Develop context engineering strategies, dynamic context assembly, and GraphRAG-based retrieval
- Address generalization and adversarial robustness across 200+ languages and strategies
- Contribute to evaluation and deployment strategies for large-scale AI moderation systems
- Collaborate with cross-functional teams to translate research into scalable solutions
Key requirements- PhD in Computer Science, Data Science, AI, or related field (or close to completion)
- Strong understanding of cutting-edge LLM research and practical implementation experience
- Proficiency in Python, Rust, or C++ and deep learning frameworks (PyTorch, DeepSpeed, Megatron, vLLM)
- Experience with distributed computing, training/fine-tuning/inference optimization, and PEFT, RL, MoE, CoT or Langchain is a plus
- Excellent problem-solving
- Creative mindset
- Ability to drive research from idea to implementation
- Python
- Rust
- C++
