x
Software Engineer - Cloud Infrastructure, Big Data and Web Full-stack Development.

Yufei(Kenny) Zheng

  • Technical Background
    Undergraduate Study
    • Bachelor of Engineering in Computer Science, GPA: 3.75/4.0, Graduated with distinction
    • CourseWork: Data Structure, Operating System, Database Priciples, Computer Networks, Advanced Computer Architecture, Compiler Principles...
    • Research Experience: Longyan Fruit Mall project, supervised by Prof.Gaoshou Zhai. Undergraduate project: 3D Human Pose Estimation System, supervised by Prof. Yongqi Sun

    Graduate Study
    • Master of Science in Software Engineering, GPA: 3.63/4.0
    • CourseWork: Introduction to Computer System 18613(A), Cloud Computing 15619(A), Foundation of Software Engineering(18652), How to Write Fast Code I/II(18645, 18646)
    • 6 Teamwork experience on course projects, ALL reveived top ratings from my teammates in CATME tools

    Technical Skills

    Strong background knowledge in Cloud Infrastructure(Kubernetes, Istio, AWS), Big Data(Spark, Snowflake, Data pipeline), High Performance Computing(SIMD, OpenMP and CUDA) and Full Stack Web Development(Frontend: React, Vanilla JavaScript Backend: Node.js, SprintBoot and DJango). He has three internship experience in HonestCoin(2019 Java software Developer), data.ai(Big Data Engineer) and Verkada(Infrastructure Team).

  • Projects Overview
    Cloud Computing

    15619 Team Project(Cpp(Drogon), AWS RDS(MySQL), EKS with Managed Node Groups, Kubernetes, Docker, SparkSQL)

    - QR Code en(de)code application, BitCoin Transaction ledger application(CPU Intensive), Twitter recommendation application(ETL, RDS query, memory intensive application)

    Uber Message Recommendation(Java, Kafaka&Samza)

    - Streaming data processing to recommend order for drivers

    Parallel Computing

    Fast CNN(CUDA, OpenMP)

    - Reimplemented all layers of CNN with CUDA, OpenMP and profiling performance by specifying different layer sizes

    Fast Canny Edge detection

    - Implemented four kernels with embeded Assembly Language, SIMD and OpenMP, achieved 90%+ theoretical peak performance.

    Web Applications

    Demo videos are available at below sections

    Emergency Social Network(Python, Django, TypeScript, React.js, MongoDB) [https://esn-yufei.herokuapp.com]

    BookHub(JavaScript, Node.js, Vue.js, MongoDB) [http://www.yufeizheng1998.com:8087/#/login]

    Incident Response Application(TypeScript, React.js, Node.js, MongoDB, CircleCI) [https://s22-sem-ir.herokuapp.com/]

    Longyan Fruit Mall(Java, Spring, SpringMVC, Mybatis, Redis, ElasticSearch, Node.js, Vue.js)

    High Performance Computing

    High Performance Canny Edge Detection - CPU version(Cpp, SIMD, OpenMP)

    CUDA Accelerated CNN - GPU version(Cpp, CUDA(Shared Memory, Pinned Memory, Constant Memory and Thrust))

    Artificial Intelligence

    3D Human Pose Estimation System(Graduation Project) - CNN, GAN, Semi-Supervised Learning

    Other Meaningful Projects

    CMU 18613(Introduction to Computer System) Labs Bundle

    Distributed Golang Crawler(Golang, ElasticSearch)

    For more detail information, please refer to "Selected Projects" Part to see the demo video or related material...
  • Internship Companies

    Verkada - Data Platform and Infrastructure team

    (Got the return offer)top start-up companies rated by Linkedin

    My main project is to support multi-arch features for the CI/CD systems.

    • Launched Graviton Infrastructure from scratch(Linux CIS ARM64 AMI, spot instance: Spot.io, reserced: ASGs)
    • Implemneted multi-arch base images for all service images with Codebuild.
    • Designed multi-arch cloud image batch-build and integrated multi-arch build with Auto-deployment system via Codebuild, 8x+ faster than local docker buildx cross-platform build.
    • Designed multi-arch MonoCI batch-build webhook, CI image cache via DynamoDB, EventBridge, Codebuild and handcrafted native local unit-tests env for M1 Mac, 10x faster compared with QEMU cross-platform build & run.
    • Implemented AWS backup resources validation for cross-region Disaster Recovery with boto3, AWS Lambda, IAM assume-role and Cloudwatch Event


    data.ai(Former App Annie) - Data Tech team

    App Annie is a unicorn company in Mobile APP analysis company, it devotes to become the best APP data provider in the world. We have 12 global Offices in US(San Francisco HQ), China, Canada,England, NetherLands, Singapore.. I was interned at the core engineering team - Data Tech.

    The project I was building from 2020 - 2021 is Unified Data Warehouse.

    • Data Platform updated from 1.0 to 2.0 to enable more flexible scheduling, processing pipeline and utilization of resources managed by Kubernetes
    • Migrated Rating historical data(Trillion Records) to Snowflake and developed new routine ETL job to achieve higher data quality.
    • Snowflake Data Delivery System: Collect Schema and create Snowflake View for customers instead of Bulk Sharing(My Rating Dataset is shared with the Biggest Customer Tiger Global Manament)
    • Developed Unified Data Warehouse Reader to Pre-optimize the AQL(Like GraphQL in Meta) and redirect query to Snowflake or Traditional DB(make our data warehouse become both OLAP and OLTP system).

    HonestCoin(Volunteer Internship)
    • Developed wheel for company to manage runtime applciation log(Java, SpringBoot, Dubbo, Redis, Docker, Maven)
    • Implemented Transaction Processing Application(Java, RabbitMQ, Spring, Dubbo)

  • 6+

    Programming Language

  • 10+

    Teamwork
    Projects
Portfolio

Selected Projects

CMU18653 -Software Design and Architecture

Team 4

CMU18652 - Foundation of Software Engineering

Team 4

High Performance Canny Edge Detection

CMU 18645 - How to Write Fast Code II

Canny Edge Detection is one of the most well-known algorithms developed by John F. Canny to detect a wide range of edges in images. This algorithm can find the edge precisely. However, the algorithm is inefficient when processing an image on CPUs, even though each pixel is processed independently. To address this issue, paralleling the image processing on each pixel can enhance the algorithm’s performance.

In this project, we will try to optimize the performance of Canny Edge Detection by using SIMD instructions and OpenMP to process pixels simultaneously.

For the details of the design or some snippets of the code, please email me. I can share some useful thoughts.

  • Team Members Changning Tsai, Hailey Pan, Yufei Zheng
  • Paper Link READ

3D Human Pose Estimation System

BJTU Undergraduate Project
Enable to lift 3D Human skeleton pose from RGB 2D image.
  • Unsupervised learning: It's not even necessary for you to feed this model 2D labeled data, just the 2D joints will get all net worked.
  • 3D data FREE: You don't even need the 3D data to train your model, by firstly generate the 3D pose and secondly generate the 3D pose by transformation and inverse transformation, then get MPJPE loss from these two 3D pose will eliminate the scarecity of 3D data.
  • Flexibility: Someone might wonder that what if I got some 3D pose data? Congratulations! You can absolutely firstly compute the 3D regression loss to get the model "warm start", that will help a lot.
  • Avoid OverfittingBy using discriminator to fit the normal distribution of normal human pose and geometrical constraints of human limb symmetry, some traits learned by the model can be robust to multiple types of data.

AVL based Malloc and Concurrent Network Proxy

CMU 18613-Coursework
AVL Malloc Memory Allocator(Implemented From Scratch with C)
  • Implemented a Dynamic Memory Management Unit(MMU) to achieve 77.3% average space utilization with high throughput, ranked top 1% out of 300 CMU students.
  • Keep tracking free blocks by segregated lists and with optimized greedy best search policy to increase space utilization without much lower down throughput.
  • Optimized the throughput of best search policy by implementing AVL and trid dozens of constant parameters to find a balance between space and time utilization.

VPN(with Cache, Concurrency supported)
  • Implemented Safe I/O libraries using C and Linux POSIX, included safe File IO, scanf, printf and etc.
  • Handcrafted the Network communication frameworks by using Linux Socket and Mutex Linux
  • Designed the LRU Cache from scrach for Network Requests and passed the CMU robustness and throughput tests.
  • Implemented concurrency features for the proxy service, introduced read_write_lock and mutex_lock to avoid race and increase the proxy's throughput.
  • Student Yufei Zheng
  • Course Website SEE

Fast CNN

CMU 18646 - How to Write Fast Code II

GPU & CPU Accelerated CNN

  • Tech Lead Yufei Zheng
  • Designed Multi-core and Many-core acceleration for CNN training via Cpp, OpenMP, CUDA and Thrust.
  • Implemented GPU parallel patterns for four kernels and accelerated by using constant and shared memory.
  • Leveraged multiple cores in CPU via OpenMP and benchmarked profiling for CPU and GPU performance.

Twitter Analytics Web Services

CMU 15619 - Cloud Computing

High Performance Web Services

  • Data Infra Lead Yufei Zheng
  • Implemented two CPU-intensive and one Memory-intensive web services via Drogon.core(Cpp), Jooby(Java).
  • Designed Key-Value style schema in MySQL and developed ETL logic and data lake to process 1TB Twitter dataset via \textbf{Scala, SparkSQL} by using Google Cloud Dataproc.
  • Launched AWS RDS, EKS and Load Balancer with \textbf{Terraform} and managed cloud services by \textbf{Kubernetes, Helm}.

To be continue

TODO
Skills

Programming Skills

  • 90% - Development experience in large scale projects(Like internship, research). Read related books and blogs to get deeper usage. Solved hundreds of problems in Leetcode.
  • 75% - Development experience in side projects(Like Udemy Project, medium workload course project). Know the special characteristic of this language.
  • 60% - Development experience in small projects(certain tasks of course projects). Can easily ramp up when refering new development frameworks.
Python90%
Golang90%
C++90%
SQL90%
Java75%
TypeScript75%
Scala60%
Assembly Language60%
Internships

Working Experience

  • May.2022 - Aug.2022
    Software Engineer Intern - Infrastructure Team
    Verkada - San Mateo, CA
  • Oct.2020 - Mar.2021
    Software Engineer Intern - Data Tech Team
    App Annie - Beijing, CN
  • Aug.2019 - Nov, 2019
    Software Engineer Intern(Non-paid)
    HonestCoin - Beijing, CN
News

Latest News

September 02, 2021

To be updated

Waiting to be filled up...

August 17, 2021

To be updated

Waiting to be filled up...

July 05, 2021

To be updated

Waiting to be filled up...

April 22, 2021

To be updated

Waiting to be filled up...

Contact

Get in Touch

Email is the my preferred way to get contacted..

  • Newark 94560, CA, USA
  • yufei.zheng1998@outlook.com
  • http://www.github.com/YufeiZheng1998

晋ICP备2022001350号-1