Yuangang's Website

ABLATOR: Robust Horizontal-Scaling of Machine Learning Ablation Experiments

Python, MLOps, Docker, Github Action, Pytorch, Pytest

AutoML,2023
Github / Paper

A distributed experiment execution framework for ablation studies. ABLATOR provides a wrapper for your model and a Trainer class for you to prototype on your method and scale to thousands of experimental trials with 1 code change.

Automates analysis artifacts, cutting down manual review processes.

Integrates comprehensive functionalities (HPO, training, tuning, analysis) outdoing counterparts like Ray, Optuna, etc.

Integration with established libraries (e.g., PyTorch Lighting), ensuring versatility without complexity.

Reduces dev-ops burdens, redirecting focus to innovative aspects of ML research.

RocketMqOperator: Auto Mangane RocketMq Cluster Instance on an Cloud Platform

Kubernetes, Docker, Go, CI/CD, CRD, Operator-SDK, Helm3, Prometheus, Grafana

SenseTime (The first listed AI company in China), 2023
Video Introduction

The RocketMQ Operator automatically deploys and manages RocketMQ clusters on the Kubernetes-based cloud environment. Apache RocketMQ is a popular distributed messaging and streaming platform.
Support Features:

Horizontal Scaling - Safely and seamlessly scale up each component of RocketMQ.

Rolling Update - Gracefully perform rolling updates in order with no downtime.

Multi-cluster Support - Users can deploy and manage multiple RocketMQ name server clusters and broker clusters on a single Kubernetes cluster using RocketMQ Operator.

WebRR: Automatic testing tool based on record/replay for web application cross-browser compatibility

Docker, Node.js, JSON, Vue.js, RobotFramework

Chinese Academy of Sciences (Top Research Institute), 2021-2022
Paper

The tool features a low-code approach, allowing users without extensive programming skills to easily create and deploy complex automation tests.

Utilized a low-code model, simplifying test creation for users with limited programming background.

Ensured cross-platform testing, enhancing application consistency across diverse environments.

Automated script generation from user interactions to JSON, then to Robot Framework formats.

Deployed in key enterprises like Southern Power Grid, streamlining their testing processes and reducing errors.

Broadened testing participation due to the tool's ease of use, boosting software quality and user satisfaction.

College professional development analysis platform

Full Stack Website, Web crawler, Python, MongoDB, MySQL, Java, Spring, Spring MVC, MyBatis, React, Docker

National-level Innovation Training Program, 2019
link

Led a team to collect and analyze job recruitment data using Web crawler and Python libraries, aiding university major and curriculum planning.

Stored data using MongoDB and MySQL databases for subsequent data processing and analysis.

Built data visualization interfaces using Spring and React technologies, improving data readability.

Used Docker for containerization and CI/CD technologies for automated operations, enhancing the efficiency of project deployment and operation, and ensuring project stability and reliability.

Achieved recognition as a "National Level Innovative Excellence Project for College Students" and adopted by school for professional development data support.

EventMaster

Full Stack Website, Node.js, AngularJS, GCP, AJAX

USC CSCI571, 2023
Github

Developed a full-stack event search platform using JavaScript/Node.js, Express, AngularJS, TypeScript, and Bootstrap, aggregating event information and enhancing user interaction.

Integrated APIs such as Ticketmaster, Spotify, Google Maps, and social media to enrich data display on the frontend.

Deployed the project on the Google Cloud Platform(App Engine) with performance optimization.

Yuangang Li