Dezhi Yu

Staff Engineer

Tiktok Inc.

Summary

Dezhi Yu is a professional developer who focuses on Cloud Native now. He has strong knowledge of Go, Objective-C and network programming. With these skills, he created quite a few widely used applications and frameworks in the company. He also joins some organizations, becoming professional Membership of ACM / IEEE / IEEE-CS / CCF. He writes his personal thoughts on programming and tech in his personal blog that accumulates 7.93 million PV and 4.43 million UV. He is active on Github, which Github world stars rank 149 as well as world followers rank 91. He loves algorithms and advanced mathematics, winning medals in ACM-ICPC and Advanced Mathematics Competition.

Dezhi Yu also is a globetrotter, traveled around the world before the age of 28, and his footprints will soon cover seven continents (coming soon). He is extreme sports enthusiasts, like skydiving, skiing, diving, hot air ballooning, helicopters. He is always willing to try new things, and keeping to learn from them.

Now, he works for TikTok Inc as a Staff Engineer.

Interests

  • Cloud Native
  • Artificial Intelligence
  • Mobile Development
  • Computational Linguistics

Education

  • MS in Computer Science, GPA 4.0/4.0

    Stanford University, CA, United States of America

  • MS in Software Engineering, GPA 3.95/4.0

    Carnegie Mellon University, CA, United States of America

  • B.Eng in Computer Science, GPA 3.8/4.0

    Huzhou University, Zhejiang, China

Experience

 
 
 
 
 

Staff Engineer - Cloud Infrastructure

Tiktok Inc

Feb 2023 – Present Mountain View, California, United States
tiktok
 
 
 
 
 

Quantitative Algo Developer - Smart Contract (DeFi 2.0)

Aperture Finance

May 2022 – Present Mountain View, California, United States
tiktok
  • Led Solana Projects as a Tech Owner, designed and implemented Pseudo-Delta Neutral Hedging Strategy which uses Leveraged LP farming protocols in Solana, using Francium, Tulip.
  • Extracting risk factors conditioned on solana volatile asset characteristics with an autoencoder at daily frequency and create synthetic training data using GANs, using historical data for the past 5 years.
  • Denoise dataset using wavelets and the Kalman filter and combine e Zipline offline and on Quantopian to test individual and multiple alpha factors using NumPy, pandas, and TA-Lib. Utilized Python libraries backtrader and Zipline to conduct backtesting on ML-driven strategies in historical market environments, reducing the probability of putting financial resources at risk when deploying candidate smart contracts in real-time markets by 30%.
  • Developed smart contracts using Rust, which employ Alpha factors-generated signals to dynamically adjust long and short positions for optimizing portfolios, accumulated a net income of $2 million in one month on the Solana chain.
 
 
 
 
 

Staff Engineer

Binance.com

May 2021 – Present Singapore City, Singapore
alibaba
  • Led a team consisting of 8 engineers to implement from scratch and successfully launch the firm’s first Strategy Distribution Engine – Themis, a global smart strategy engine based on traffic flow and predicate conditions’ strategy.
  • Simulated ads data on the Strategy Distribution Engine to do offline data analysis and queried and fed production data to the system and preformed maximum 1K online A/B experiments simultaneously and release a maximum of 20K Strategy per day.
  • Themis ecosystem combines and compares online and offline data for enhanced resource allocation strategy with accumulated 200 million users, facilitating transaction volumes reaching $1000 million per week.
 
 
 
 
 

Senior Back End Engineer |Senior Research And Development Engineer

AliCloud

Apr 2018 – May 2021 Shanghai, China
alibaba
  • Participated in Tmall Double Eleven online shopping promotion for 2 consecutive years; Led team consisting of 5 engineers.
  • Designed and implemented a ads targeting system, which utilizes user’s historical purchase records and recent search history to recommend promotional deals based on customer profiling behavior data, dynamically adjusting the intensity of discounts in deals based on users' recent engagement levels data, resulting in an annual revenue growth exceeding 48% and an overall purchase rate improvement of 8%, helped company create a new transaction record of 268.4 billion RMB a day in 2019.
  • Optimized on-line and off-line recommendation model for advertisements, improved advertisement push, banners, and SMS performance by 25%; Wrote HiveQL and SQL to gather original data, preprocessed data set by data cleaning, categorical feature transformation and normalization; Enriched user-profile features and optimized user-profile update strategies to provide reliable features for machine learning models; Trained XGboost and LR model, increased AUC from 0.59 to 0.77. Did data resampling and undersampleing for imbalanced data, imputed missing values, evaluated model by recall and precision.
  • Taco ecosystem successfully handles 1100 million+ push notifications per day, with delivery rate above 97% and delivery time of 0.72-1.2 seconds; daily peak value can reach 750k online connections with 30k-50k QPS.
 
 
 
 
 

Senior Software Engineer

ele.me

Feb 2017 – Apr 2018 Shanghai, China
eleme
 
 
 
 
 

Infrastructure Team Leader

Shanghai Fangchuang Financial Information Service Co., Ltd.

May 2016 – Feb 2017 Shanghai, China
fangchuang
  • Led engineer team to designed and implement a FaaS infrastructure snapshot system using Java. Speed up the average response time by 65% and improved the service robustness by 81% via parallelization and local caching techniques for read-heavy operations with PB level daily caching volume.
  • Migrate 12 backend docker service to AliCloud, a cloud kubernetes platform and refactored the infrastructure by using redis and mysql in AliCloud, which increased the stability of the company’s entire back-end infrastructure by 65%.
  • Designed and implemented a internal system to intelligently match projects for investors, utilizing a recommendation system model, text abstraction model, and Text-to-Speech model, increasing the success rate of project bidding by 40%.
 
 
 
 
 

Infrastructure Team Leader

Shanghai Ping An Smart Technology Co., Ltd.

Nov 2015 – Apr 2016 Shanghai, China
yqb
  • Designed and implemented the 1QB Wallet recommendation system, matching clients and products based on the keywords of previously searched items to boost sales, using RPC and HTTP framework in Java, improving the accuracy by 30%.
  • Building new Code Review platform from scratch; developed Java package deployment and remote backup distribution platform, including Jenkins integrated static code scanning on daily basis and CI/CD pipeline, increasing efficiency by 5 times..
  • Developed a trading system in Go, leveraging the Kafka Messaging System, to enable users' participation in flash deal ordering events, enabling millions of users to engage in millisecond-level promotional activities. In the first year since the release and launch of this system, it has contributed an incremental revenue of 10 Billion RMB in GMV to the company.
  • Refactored the task queue and task manager using Celery, RabbitMQ and MySQL to support larger-scale, distributed tasks; Queue latency reduced by 20%, Consume rate increased by 4x and Memory used decreased by 40%.
 
 
 
 
 

Software Engineer

Quatanium Technology Co., Ltd.

Jun 2013 – Nov 2015 Shanghai, China
quatanium
  • Designed and implemented firm’s first Cluster Observability Platform from scratch. Built a log service based on logstash, reformat, and transfer logs from grouping system to Kibana log platform, handing million logs per day with very high stability and availability which annual availability rate reaches 99.999%; Created Grafana dashboards to report the counters of logs.
  • Designed and implemented CI/CD pipelines with YAML and deployed Docker containers to AWS, making them automatically build, pack, and deploy the log service to production, serving an average of 200 integrations per day.
  • Designed and implemented an alert service for error logs, which can report the errors and create fix tickets for DevOps engineers automatically, reducing fix waiting time by 80%.

Accomplish­ments

Preparing for Google Cloud Certification: Cloud Architect Specialization

Google-Cloud
See certificate

Architecting with Google Compute Engine Specialization

Google-Cloud
See certificate

Architecting with Google Kubernetes Engine Specialization

Google-Cloud
See certificate

Programming with Google Go

UCI
See certificate

IEEE Professional Membership

IEEE
See certificate

IEEE-CS Professional Membership

IEEE-CS
See certificate

CCF Professional Membership

CCF

The 2016 Best Recommended Author of jianshu

jianshu
See certificate

The 10 Best Original Authors of 2016 juejin.im

juejin
See certificate

National Computer Rank Examination (Level 4 Network Engineer)

NCRE
See certificate

The 36th ACM-ICPC Asia Regional Contest Shanghai Site 2011 Fudan University - Honorable Mention

ACM-ICPC
See certificate

The 8th Zhejiang Province University Programming Contest - Bronze Medal

ACM-ICPC
See certificate

The ACM-ICPC Asia Regional Google Cup 2011 ACM-ICPC China Shanghai Fudan Invitational Programming Contest Fudan University - Bronze Medal

ACM-ICPC
See certificate
*

Themis

Themis is the first of Strategy Distribution Engine systems in the company, and have serviced 200 million user. Themis, a smart strategy engine based on traffic flow and predicate conditions’ strategy, which provides handy service for vendors and consumers.

Prometheus

Prometheus Ghost theme is clean and very modern. The design is highly focused on typography. But there is also the perfect combination of image and white space in this design. This theme is perfect for a personal and professional blog.

Taco

Taco is one of the fast golang-based message push systems in the company, and have serviced 300 million user. Taco provides multiple push methods such as pushing a single user, pushing a group of users in batches, and accurately pushing specific user tags. The low-latency feature greatly empowers the logistics rider’s message reach scenario.

threes-ai

Deep Reinforcement Learning for the Threes! game. website: https://threes.halfrost.com/

Stormeye

Stormeye is the company’s internal project and it is the basic weather service.

Tank War AI

This project is for 2017 year ele.me’s Hackathon Programmer Contest. The competition requires designing a AI tank. The two sides of the battle tanks play in the sandbox. The competition requires submitting a working docker image. So this is my solution.

Clairvoyant

Clairvoyant is the company’s internal project and it is the basic service for collecting and analyzing rider trace.

LeetCode-Go

The book about solutions to LeetCode in Go, 100% test coverage, runtime beats 100%. The book has won the praise of many people from CMU PhD/Google/Apple/Microsoft/Facebook/Amazon. Since this book released on 13, August, 2020, it had occupied the Github trending top 10 for 2 consecutive months. It has 0.74 million Words, 2 month cumulative 1.26 million Page Views, 0.56 million Unique Visitors, 68K PDF downloads.

LPDTeam

LPDTeam APP is a delivery app tailored for delivery personnel by eleme Takeaway. It serves a large number of franchisees and eleme’s self-operated delivery team. It aims to help distributors better and faster serve eleme takeaway merchants and customers Provide delivery services.

eleme crowdsource

Hummingbird crowdsourcing is waiting for you to join! Create the concept of national distribution, everyone can become a rider and start the distribution life. Produced by Ele.me, the country’s largest online ordering platform, a large number of waybills are softened.

vue-objccn

Use Vue.js to develop a cross-platform full stack application of Objc china.

Youcai

Ele.me’s O2O food supply platform provides restaurants with high-quality food delivery. 2000 fresh food suppliers covering dozens of cities across the country, direct delivery will arrive next day! Direct supply from manufacturers, direct delivery from well-known flagship stores, and free home delivery from surrounding vegetable vendors.

Halfrost-Field

This repo concludes my technology blog、speech slides、electronic publication.

iOS-Weekly

iOS-Weekly are not just porters of information, but also evangelists of value. The content is not limited to specific content related to iOS technology. It also includes information that we believe is valuable to iOS developers, such as software engineering and programmer topics.

Fangke

Fangke APP is a mobile investment and financing platform launched by Fangchuang. Committed to providing investors and entrepreneurs with an efficient, fast and convenient mobile investment banking experience. Recommend only high-growth developmental companies, which is convenient for investors to screen and compare.

PingAn YQB

YQB is a one-stop financial life shopping platform under Ping An of China, committed to creating value for every point of users. It covers many areas such as point service, current financial management, regular financial management, credit services, shopping, convenience of life, transfer and repayment, financial services, etc.

Qhome

Trying to make your home awesome? Download the most innovative and user-friendly smart home app - Qhome by Quatanium Technology Co., Ltd. You will discover that your home could be more intelligent than you ever expected. Everything is awesome!

Chinese Dictionary 10001

A Chinese dictionary with more than 20,000 Chinese characters and detailed explanations, a collection of phonetic transcription, pinyin, stroke count, extra-part strokes, and interpretation. It also supports powerful search methods such as Chinese characters, pinyin, radicals, and stroke counts.

Smart Sokoban

This application is mainly to realize the sokoban smart game based on Android. Sokoban smart game(https://www.sokoban.jp/) is an ancient game from Japan. Its design purpose is to train people’s logical thinking ability. The game scene is usually set in a small space warehouse, and it is required to place the boxes in the designated location.

Skills

Golang

80%

Docker

60%

MySQL

80%

PostgreSQL

70%

Nginx

70%

RabbitMQ

70%

Redis

70%

Jenkins

60%

JavaScript

60%

HTML5

60%

CSS3

60%

Vue

60%

Weex

60%

Electron

60%

Apple

90%

Adobe Illustrator

90%

Adobe Photoshop

90%

Sketch

90%

Git

80%

Gitlab

80%

Phabricator

80%

Contact