Yu Gong YG

Yu Gong 龚禹

Director of Engineering @ TikTok · AI Search · LLM Agents · Recommender Systems

Greater Seattle Area

Google Scholar GitHub LinkedIn X / Twitter

About

I'm a Director of Engineering at TikTok, leading a 40+ person team building AI Search and Shopping Agent for global e-commerce. I have 9+ years of experience in large-scale search and recommender systems, previously at ByteDance and Alibaba.

My work spans search, recommendation, and agent systems, which I view as different forms of large-scale decision systems for resolving user intent under uncertainty — from retrieval and ranking to multi-step reasoning and decision making.

My current research interests include LLM-powered search, LLM agents (post-training, personalization, and evaluation), and recommender systems. Recipient of the SIGIR 2017 Best Paper Honorable Mention (IRGAN, 800+ citations).

News

Experience

2024 – Present

Director of Engineering TikTok

Lead a 40+ engineer organization owning AI Search and Shopping Agent systems for TikTok global e-commerce across multiple regions (US, Europe, Southeast Asia and more), driving end-to-end product and technical roadmap for search systems serving ~140M daily e-commerce queries in the US alone.

LLM-powered Search

Led the evolution of search systems toward LLM-based retrieval and ranking:

  • Query Planning — LLM multi-agent framework (Analyzer-Identifier-Rewriter-Judger) for resolving ambiguous queries into latent product intents.
  • Generative Retrieval — decode-only generative models for product retrieval.
  • Large Search Model — Transformer-based search model with parameter and behavior sequence scaling up across retrieval and ranking.
  • Multi-Modality — multi-modal representation fusion and co-training for retrieval and ranking.

Shopping Agent

Built and launched TikTok's Shopping Agent from scratch:

  • Agent Framework — Planner-Executor-Verifier architecture (extended ReAct) with multi-turn context management and heterogeneous tool use (Web / Product / Video Search).
  • Memory & Personalization — Memory-as-Tool design that leverages user behavioral signals for personalized responses.
  • Post-Training — rejection-sampling SFT with synthetic multi-turn data for model cold-start; RL fine-tuning with Rubric Reward & Deep Search Reward. Research on Active Reasoning of LLM Agents → T3 & AREW.
  • Evaluation & Benchmark — built ShoppingBench (deep product search), APeB (personalized agent behavior), and rubric-based report evaluation.
2021 – 2024

Senior Engineering Manager / Tech Lead ByteDance

Led a 20+ engineer team building Douyin Mall's core recommendation and product growth systems from 0→1, defining the foundation for personalized e-commerce and contributing to its growth into China's Top 3 largest e-commerce platform by GMV. Owned end-to-end recommendation architecture supporting ~170M daily active users across products, short videos, and livestreams, spanning retrieval, ranking, and cold-start systems.

Homepage Feed Recommendation

Built and scaled the full-stack recommendation pipeline for Douyin Mall's homepage feed:

  • Deep Retrieval — Beyond-dual-tower retrieval with cross-tower feature interaction, multi-objective cascade for full-pipeline consistency with advanced loss optimization, and advanced i2i modeling.
  • Large Ranking Model — Pioneered Large Ranking Model with parameter scaling up across sparse and dense components, long-sequence-to-graph modeling, and scaling laws exploration.

Product Cold-Start System

Led the design of a dedicated cold-start system to bootstrap new products at scale:

  • Traffic Strategy — Built a lane-separated serving architecture decoupled from the main traffic flow, enabling controlled exploration of new products; developed uplift modeling for personalized traffic allocation, balancing exploration efficiency and conversion impact.
  • Efficiency Optimization — Enhanced cold-start ranking with multi-modal representations, similar-product knowledge transfer, and cold-start-aware training strategies under sparse feedback.
2017 – 2021

Tech Lead Alibaba Group

Pioneered EdgeRec — the industry's first on-device AI recommender systems in a billion-scale commerce feed for Mobile Taobao. Invented the Generator-Evaluator two-stage re-ranking framework (GRN, GE), now the standard paradigm for listwise re-ranking. Created Alibaba's Edge-X-Platform (Edge-Cloud MLOps), powering 10+ business lines across Alibaba.

Selected Publications

Total citations: 1,690 · h-index: 16 · Full list on Google Scholar

LLM Agents

  • On Information Self-Locking in Reinforcement Learning for Active Reasoning of LLM Agents D Zou, Y Chen, F Feng, M Li, P Li, Y Gong, J Cheng Submitted to ICML 2026 Project Lead
  • Reducing Belief Deviation in Reinforcement Learning for Active Reasoning of LLM Agents D Zou, Y Chen, J Wang, G Yang, M Li, Q Da, J Cheng, P Li, Y Gong ICLR 2026 Oral Project Lead Cited by 1
  • MemRerank: Preference Memory for Personalized Product Reranking Y Gong, et al. arXiv 2026 Project Lead

Benchmarks

  • APeB: Benchmarking Personalization Ability of Large Language Model Agents Y Gong, et al. Submitted to ICML 2026 Project Lead

Recommender Systems

  • EdgeRec: Recommender System on Edge in Mobile Taobao Y Gong, Z Jiang, Y Feng, B Hu, K Zhao, Q Liu, W Ou CIKM 2020 First Author Cited by 110
  • Personalized Adaptive Meta Learning for Cold-start User Preference Prediction R Yu, Y Gong, X He, Y Zhu, Q Liu, W Ou, B An AAAI 2021 Project Lead Cited by 101
  • Exact-k Recommendation via Maximal Clique Optimization Y Gong, Y Zhu, L Duan, Q Liu, Z Guan, F Sun, W Ou, KQ Zhu SIGKDD 2019 First Author Cited by 63
  • GRN: Generative Rerank Network for Context-wise Recommendation Y Feng, B Hu, Y Gong, F Sun, Q Liu, W Ou arXiv 2021 Project Lead Cited by 32
  • Revisit Recommender System in the Permutation Prospective Y Feng, Y Gong, F Sun, Q Liu, W Ou arXiv 2021 Project Lead Cited by 31
  • Query-based Interactive Recommendation by Meta-path and Adapted Attention-GRU Y Zhu, Y Gong, Q Liu, Y Ma, W Ou, J Zhu, B Wang, Z Guan, D Cai CIKM 2019 Co-First Author Cited by 21
  • Gift: Graph-guided Feature Transfer for Cold-start Video Click-through Rate Prediction Y Cao, S Hu, Y Gong, Z Li, Y Yang, Q Liu, S Ji CIKM 2022 Cited by 18

Information Retrieval & Search

  • IRGAN: A Minimax Game for Unifying Generative and Discriminative Information Retrieval Models J Wang, L Yu, W Zhang, Y Gong, Y Xu, B Wang, Z Peng, D Zhang ACM SIGIR 2017 Best Paper Honorable Mention First Industry Author Cited by 829
  • Conceptualize and Infer User Needs in E-commerce X Luo, Y Yang, KQ Zhu, Y Gong, K Yang CIKM 2019 Cited by 21
  • Query Tracking for E-commerce Conversational Search: A Machine Comprehension Perspective Y Yang, Y Gong, X Chen CIKM 2018 Cited by 13
  • A Minimax Game for Instance Based Selective Transfer Learning B Wang, M Qiu, X Wang, Y Li, Y Gong, X Zeng, J Huang, B Zheng, D Cai, ... SIGKDD 2019 Cited by 60

NLP & Text Generation

  • Deep Cascade Multi-task Learning for Slot Filling in Online Shopping Assistant Y Gong, X Luo, Y Zhu, W Ou, Z Li, M Zhu, KQ Zhu, L Duan, X Chen AAAI 2019 First Author Cited by 42
  • Multi-Modal Generative Adversarial Network for Short Product Title Generation in Mobile E-Commerce J Zhang, P Zou, Z Li, Y Wan, X Pan, Y Gong, PS Yu NAACL 2019 Cited by 35
  • Automatic Generation of Chinese Short Product Titles for Mobile Display Y Gong, X Luo, KQ Zhu, W Ou, Z Li, L Duan AAAI 2019 First Author Cited by 35
  • Representing Verbs as Argument Concepts Y Gong, K Zhao, KQ Zhu AAAI 2016 First Author Cited by 20

Education

2014 – 2017

M.S. in Computer Science Shanghai Jiao Tong University

Research Assistant in ADAPT Lab. Research areas: Information Retrieval, NLP, Machine Learning.

2010 – 2014

B.S. in Computer Software Engineering Xi'an Jiaotong University

Google Excellent Scholarship recipient (awarded to 100 students in China).

Honors & Awards