Ziyao Chen

Tech Lead, Weixin (WeChat) Search — Weixin AI, Tencent
Guangzhou, China

Biography

I work on web-scale search at Weixin AI, Tencent, where I lead the offline content-understanding, retrieval and pre-ranking stack behind Weixin Search — the search engine serving WeChat's user base. My interests sit at the intersection of large language models and search systems: how to represent and retrieve content across tens of billions of documents, and how to make LLMs genuinely useful inside a production ranking funnel rather than just in a demo.

I started at Tencent as an intern in SNG in 2018 and joined Tencent Cloud full-time in 2019, back when conversational AI was the thing everyone was building. I spent that period shipping chatbots — open-domain chit-chat, FAQ retrieval, and KGQA over structured knowledge. In 2020 I moved internally to WeChat Mini Program Search, working on ranking and on squeezing BERT into production latency budgets with TensorRT. When Tencent consolidated its search efforts into Weixin Search in 2021, my team came along — and we won the internal bake-off against several competing engines. That system is what powers WeChat search today.

Since then I have rebuilt the offline content-understanding and retrieval/pre-ranking architecture for both AI Search and main search from the ground up; the version running in production today is that rewrite. I have also led re-architectures of several vertical searches — image search, text-selection search and comment search — and work closely with the fine-ranking and relevance stack.

Research Interests

Large Language Models Search & Information Retrieval Content Understanding Retrieval & Pre-Ranking Multimodal Learning Inference Acceleration

Work Experience

• 2021 — Present
  Weixin (WeChat) Search · Weixin AI, Tencent

  Offline data pipelines, content understanding, retrieval and pre-ranking. Rebuilt the offline understanding + retrieval/pre-ranking architecture for AI Search and main search from zero; that rewrite is the version in production. Also re-architected image search, text-selection search and comment search. Our engine won the company-wide internal evaluation against multiple alternatives and became the shipped version of WeChat search.

• 2020 — 2021
  WeChat Mini Program Search · Tencent

  Fine-ranking for Mini Program search, plus BERT inference acceleration with TensorRT to fit transformer models into production latency budgets.

• 2019 — 2020
  Tencent Cloud · Tencent

  Conversational AI: open-domain chit-chat, FAQ retrieval, and knowledge-graph question answering (KGQA) for enterprise chatbot products.

• 2018
  SNG (Social Network Group) · Tencent — Intern

  First taste of large-scale systems at Tencent.

Education

• 2016 — 2019
  M.Eng. in Information Engineering · Zhejiang University
• 2012 — 2016
  B.Eng. in Engineering · Zhejiang University