Tech Lead Machine Learning Engineer, TikTok Branding Ads

Build the most powerful content understanding engine on the world's leading short-video platform! Retrieve and rerank billions of videos, redefine the intelligent connection between advertising and content with multimodal large language model.



TikTok Brand Ads team is hiring multiple MLE/SWE roles, focusing on the in-depth application of multimodal large language models in brand advertising. Join us to build the most powerful content understanding engine on the world's leading short-video platform, retrieve and rerank billions of videos and redefine the intelligent connection between advertising and content with multimodal large language model!



The team is responsible for the complete technical chain from data construction, model training, offline evaluation, online deployment, inference optimization to new model exploration, covering key tasks such as multimodal semantic understanding, content matching and ranking, and cross-modal alignment.



Key technical directions:

- Multimodal-LLM model development

- Generative retrieval & ranking technology

- Ads ranking

- LLM inference opptimization



We are looking for sophisticated engineers that have strong problem solving skills and algorithm understanding to build and manage systems with high performance, scalability, and availability under complex monetization scenarios. You will have the opportunity to partner closely with a globalized engineering and product teams in a high-impact and fast-paced environment.



What you'll do:

- Lead and design key project roadmaps for an efficient large-scale video content indexing and retrieval system to support vector-level matching and filtering between ad semantics and millions of native videos.

- Use large video models to model the semantic meaning of video content and construct structured vector representations.

- Build multimodal representations of brand ads (video/picture + title/script + brand semantics, etc.) to achieve cross-modal alignment.

- Set up content understanding pipelines for various business scenarios, processing tens of millions of videos daily.

- Apply technologies such as Embedding Distillation and Hard Negative Mining to optimize the training process.

- Integrate large model inference services and design of multi-channel recall service architectures (supporting multimodal + multi-stage inference).

- Design of incremental update, cache acceleration, and offline synchronization mechanisms for multimodal pipelines.

- Drive proactive cross functional communication with PMs, Product, Data Scientist, and Designer teams, and ensure smooth ads product delivery and launching.