Generative ML Engineer - Multimodal for Recommendations

TikTok; Seattle (WA); On-site; USD 122,000 - 243,000; Full time

Job summary

TikTok is seeking a highly skilled ML/LLM researcher for the Recommendations Foundation team in Seattle. You will contribute to pre-training, mid-training, and post-training of Foundation Models, and design multimodal tokenizers to encode rich item content and improve model performance.

The role focuses on developing LLM-native recommendation systems, spanning retrieval, ranking, and end-to-end generation, with opportunities for publication and impact within a leading global platform.

Qualifications

• Bachelor's degree in CS, EE, math, statistics or related field.
• Strong foundation in machine learning and deep learning, with interest in LLMs and generative recommendation.
• Proficiency in Python and experience with deep learning frameworks such as PyTorch.
• Self-driven, with a strong research mindset and solid engineering skills.

Responsibilities

• Participate in the full training lifecycle of Recommendation Foundation Models, including pre-training, mid-training, and post-training.
• Design and train multimodal semantic tokenizers for recommendation items, leveraging multimodal foundation models to encode rich item content into discrete semantic tokens and raise the performance ceiling of Recommendation Foundation Models.
• Develop LLM-native recommendation by incorporating recommendation tasks directly into large language model training and leveraging world knowledge to improve recommendation quality.
• Build the next generation of recommendation systems powered by Recommendation Foundation Models, spanning retrieval, ranking, and end-to-end generative recommendation.

Education

Bachelor's degree in Computer Science, Electrical Engineering, Mathematics, Statistics or related discipline

Job description

TikTok is seeking a highly skilled ML/LLM researcher for the Recommendations Foundation team in Seattle. You will contribute to pre-training, mid-training, and post-training of Foundation Models, and design multimodal tokenizers to encode rich item content and improve model performance.

The role focuses on developing LLM-native recommendation systems, spanning retrieval, ranking, and end-to-end generation, with opportunities for publication and impact within a leading global platform.