JD: Lead Bing Social, Real time, part of News quality improvement. 1. Initiate and built Real time machine learning system in Microsoft. Including Real time spiking detection, real time clustering and real time ranking. Twitter Coverage 10x, Win Google: freshness 50% faster. 2. NLP with casual logic understanding. Intelligent Knowledge graph and expert middle layer semantic understand. LSAT QA competition with Google, using casual preposition i.e. "thus, so, because, therefore" etc. to build premise-conclusion sub-sentence level next token prediction pairs  in any context . Got 5 points absolutely improvement and +40% more using GPT3 signal as context relationship. 3. initiate Real Time Machine learning system in Microsoft, including real time pipeline built within secs level latency(Spiking Query 99% <7.5 secs, Twitter 95%<1.7secs), built real time spiking signals, real time clustering, real time brand new topic extraction and servicing online KV full cache to reduce all online latenc

Principal Software Eng Manager

Nearby Recommendation Project Handler | Senior Engineer (T3.3) at Tencent | Tencent Map Platform Search Center JD: Location POI service for key Tencent products (Wechat, QQ, Qzone etc.) which 1 Billion PV per day. 1. Building whole recommendation system from zero base: CTR using OpenMPI machine learning training system, ABTest system, online and offline updating circle system etc. 2. Index re-factory, traditional search index re-factory to KV(based on dense hash map) and semantic machine learning system 3 machine per set decrease to 1, resolve the recall truncation(avg 100 to 500), compute about 125K items per second, 15x better. 16 machines in 1Billion PV online Serving which Top1 POI service in China, best performance in the world. 3. Optimization: Analyze resident user and not resident user of nearby, their behaviors, different sub model according to categories, rank, distance etc. LR CTR model AUC 0.88. CTR +7.97%(from 20.04% to 28.01%), Top5 +3.07%(from 77.39% to 80.46%). Top1 in 

Basic Research, Senior Researcher(T3.3)

CTR+40% in shenma novel search recommendation

Search Exper

Twitter-like feed ads optimization 10x in 2 years.  Personalization and real time ads using recent user active behavior to simulate user query or intent. First version is 3x to traditional ads without personal and real time intent updating. Built ads platform, including parallel computing and training both OWLQN and LDA in OpenMPI2.0.  Optimize the communication in training and serving performance to top tier within 2 months.

Senior Engineer

Senior R & D Engineer

Co-Founder & CTO

LLM in coding

Know about Chewy Chen's work experience, education, bootcamps, certifications, and much more. Chewy Chen