MingYi Cui
Liaoning, China • - •-• linkedin.com/in/ming-c-/
Summary
Experienced Machine Learning Engineer with a proven track record in developing sophisticated systems across diverse industries. Proficient in designing and implementing robust models for stock trading, medical equipment detection, anomaly detection, and object classification using deep learning frameworks like PyTorch, TensorFlow, and Ray, coupled with expertise in CI/CD pipelines and cloud deployment. Skilled in leveraging AI to drive innovative solutions and possessing a strong background in full-stack development and data analysis.
EXPERIENCE
NodeAsset Oct 2021 – Jul 2024
Machine Learning engineer
Developed a deep reinforcement learning system for automated stock trading using OpenAI Gym. Defined customized actions and rewards and used Ray RLlib Algorithms (DQN, PPO, A2C, etc) based on the PyTorch framework.
Worked on Python, GO lang to scrape stock data from Interactive Broker (TWS) API and deployed to Azure Kubernetes so that trainer/inference pod can fetch data.
Designed and implemented customized Gymnasium environment to update states of a trading agent, created Azure CI/CD pipelines to build a docker image and K8 YAML files.
Developed auxiliary tools to evaluate the performance of trained models on customized data and pick up the best performance checkpoint, and save all information in MongoDB.
Johnson & JohnsonAug 2019 – Aug 2021
Data scientist/Machine Learning engineer
Developed medical equipment detection system using FastRCNN-ResNet101 backbone with over 6k classes.
Designed and implemented CI/CD pipelines on Azure to generate training data from video files, train models for different subsets of targets, and reveal inference API endpoints using the FastAPI framework.
Developed a web-based semi-automatic annotation tool to accelerate training data using an object tracking module in OpenCV and VueJS front-end framework.
Participated in molecular object detection project using HCI(high content imaging). Analyzed performance of classical ML models (CellProfiler, KS statistics, etc) and DL models (ResNet, DenseNet, GapNet, FNN, etc) and proposed modified pre/post-processing methods to enhance detection accuracy.
Worked on Jenkins pipeline and Kubernetes to train models on 6*V100 GPU cluster as VolcanoJobs, accelerate training using Pytorch multi-GPU training DDP.
Worked on patient health data analysis project that checks data completeness such as heart rate, blood pressure, etc.
Discovery Democracy LLCFeb 2018 – Jul 2019
Machine Learning Engineer
Built a Time Series Anomaly Detection system based on LSTM+VAE model to classify object categories that move on belt-conveyer using the Tensorflow framework.
Developed Python GUI applications to collect train data from 4 sensors placed on the scan area, and built around 600k time series train data. Analyzed the signal features of different types of objects and defined the features for training. Automate training data update, training, and inference with Python and shell script, achieving 91% precision.
Worked on developing an object location detection device, which detects hidden corrosion through PCMs tile, terrazzo, PRC, and non-skid surface materials. Built python GUI interface and deployed into Raspberry PI 4B model.
GoMapsFeb 2015 – Jan 2018
Full Stack Engineer
Worked on fake car insurance claims detection. Modeled users, licenses, policies, etc as graph data and developed algorithms to detect fake claims based on Neo4J GraphQL.
Developed backend APIs based on Go and PostgreSQL databases.
Maintained the company’s internal website that built on ReactJS and AWS.
Worked on analyzing money leakage issues, and sending out emails to customers using SendGrid.
Collaborated on developing car plate/driver license detection model using ALPR+Tesseract OCR.
EDUCATION
Shanghai JiaoTong University, ChinaMay 2014
Bachelor of Computer Science
SKILLS
Machine learning
Python
Django
Flask
QT
C/C++
SQL
AWS
Azure
Linux
MySQL
MongoDB
PostgreSQL
Bash Scripting
Javascript
HTML/CSS3
OpenCV
Shell/Bash
Git/Bitbucket
Docker/Kubernetes
