I work on search, ads, and recommendation systems, with a long-running focus on ranking, recommender systems, causal inference, and industrial machine learning.
Recently, I have been exploring how LLMs, generative recommendation, and agentic systems can reshape production search and recommendation: from retrieval-ranking-reranking pipelines to tool-using agents, decision traces, online evaluation, and product-facing algorithm design.
-
Agentic Search & Recommendation Building lightweight agent frameworks for search and recommendation systems, where recall, ranking, reranking, explanation, and critique can be coordinated as specialized agents.
-
LLM4Rec & Generative Recommendation Tracking and writing about industrial LLM4Rec systems, especially end-to-end generative recommendation, semantic IDs, recommendation scaling laws, and large ranking models.
-
Causal Inference for Industrial ML Studying how causal methods, bias-aware evaluation, and counterfactual thinking can make recommendation and ranking systems more reliable.
-
Knowledge Sharing Turning paper reading, engineering notes, and industrial algorithm observations into reusable public notes on Zhihu and GitHub.
-
AgenticRec A lightweight agentic framework for search and recommendation. It turns the classic recall-ranking-reranking pipeline into a council of agents, with tool orchestration, optional LLM reasoning, observable decision traces, and a built-in benchmark loop.
-
LLM4Rec-Papers Reading notes on LLM4Rec and industrial generative recommendation, including the One-series systems, semantic ID/tokenization, RL preference alignment, OneSearch/OneMall/OneLoc, and RankMixer-style industrial ranking scaling.
-
PaperNotes A curated archive of Zhihu paper notes covering recommender systems, causal inference, LLM x search/recommendation, conference paper collections, and deep-dive technical summaries.
-
kcicpt-matlab MATLAB implementation of the Kernel Conditional Independence Cluster Permutation Test and its integration with the PC algorithm for regulatory/Bayesian network structure learning.
I care about algorithm systems that are not only accurate, but also deployable, explainable, measurable, and useful in real products.
- Retrieval, matching, ranking, reranking, and recommendation strategy optimization
- User modeling, personalization, cold-start, and long-tail intent understanding
- LLM applications for search, ads, recommendation, and decision systems
- Generative recommendation, semantic item representation, and recommendation scaling
- Causal inference, unbiased evaluation, and decision-making under feedback loops
- Practical ML systems that connect research ideas with production constraints
-
Education B.S. in Computer Science and Technology, Beijing Jiaotong University, 2011-2015. Joint training in Computer Science and Technology, Peking University, 2015-2017.
-
Industry Algorithm-related work across Huawei, Alibaba Group, and Baidu, with a continuing focus on search, recommendation, ranking, and machine learning systems.
On Zhihu, I write about recommendation algorithms, search and recommendation systems, causal inference, LLM x RecSys, and applied machine learning.
As of May 22, 2026, my public Zhihu profile records:
- 10,737 followers
- 175 answers
- 39 articles
I also run the WeChat public account: 机器学习与商业智能前沿.
- AI_game: browser game experiments built with Codex
- pytorch-vdsr: PyTorch implementation of VDSR
- PRMLT: MATLAB implementations of machine learning algorithms from PRML
- Lipreading_DBN: visual speech recognition with Deep Belief Networks
- Editor-Qt: a game editor similar to RPG Maker