
Senior Research Scientist, Foundation Model (LLM/ VLLM), TikTok - Trust and Safety
- San Jose, CA
- Permanent
- Full-time
1. Lead the design, training, and deployment of foundation models, including LLM/VLM, to support a broad range of content safety tasks across modalities. Build general-purpose foundation models with centralized compute and scalable architecture, aimed at improving risk detection, compliance understanding, and moderation automation.
2. Tackle challenges in multilingual, multimodal, and low-resource scenarios by enhancing models' zero-shot and few-shot generalization across diverse safety domains.
3. Design and maintain a multimodal safety annotation framework, supporting high-quality training data and evaluation signals for both image and video understanding tasks. Explore and implement RLHF strategies to fine-tune model alignment with evolving safety policies and user intent.
4. Collaborate closely with infra, data, and platform teams to optimize large-scale training pipelines, improve model serving efficiency, and leverage centralized GPU resources effectively.
5. Partner with product, policy, and recommendation teams to integrate safety-oriented models into real-world moderation workflows, and continuously optimize for performance and interpretability.Qualifications:Minimum Qualifications:
1. Solid experience in traditional machine learning, with over 2 years of research or development experience in large models; passion for technology, and ability to dive deep into coding and debugging.
2. Strong problem-solving skills in the face of complex challenges, with a proven track record of technical breakthroughs and business impact; clear logic and sharp insight.
3. Highly responsible, self-driven, with excellent learning and communication abilities; strong interest in staying up-to-date with advances in multimodal large models.
4. A strong team player, open to exploring new technologies and driving continuous technical evolution.Preferred Qualifications:
- Publications in top conferences (e.g., NeurIPS, ICML, ICLR, CVPR, ICCV) or experience in competitive AI challenges.