Bussiness Unit:

• Corporate Risk Office

Responsibilities include:

• Form concepts and plan AI/ML projects
• Collaborate with data and software engineers to enable the deployment of AI/ML solutions across CPG’s subsidiaries
• Develop Machine Learning models to solve business problems across the company’s subsidiaries
• Advise teams on new AI/ML techniques and solutions
• Experiment with state-of-the-art AI/ML techniques to be applied to the company's subsidiaries

Bussiness Unit:

• Data Management and Analytics

Responsibilities include:

• Pioneer the development of an analytics platform, the functionalities include automatic ETL data integration process, on-the-fly Exploratory Data Analysis (EDA), automatic ML model re-training, real-time Loss Ratio forecasting dashboard for all insurance products by policy codes, automatic fraud claim detection and notification, etc.
• Pioneer the establishment of the company’s data transformation from a legacy database system to a new cloud-based data warehouse to support business intelligence (BI) activities and advanced data analytics
• Predict future claims, premiums earned, loss ratio, medical inflation, etc. according to company’s current assurance products and new businesses
• Research and develop state-of-the-art machine learning models in actuarial science field
• Implement an end-to-end ML framework from model development, testing to deployment
• Communicate insights and findings from analytic results, make recommendations to management teams to elevate the company’s business strategies

Bussiness Unit:

• Data Science and Advanced Analytics

Responsibilities include:

• Co-develop, deploy and maintain ML models closely with Data Scientists
• Customer Segmentation, KYC (Know Your Customer), clustering models
• Personalization and Recommendation models
• Demand Forecasting models
• Product Pricing and Discount Utilization models

Model development frameworks:

• PyTorch, Keras-Tensorflow. Gensim, Scikit-learn, MLFlow, etc.

Application development tool stack:

• FastAPI/Flask, Celery, Redis, Docker, etc.

Deployment tool stack (work cooperatively with DevOps):

• GitHub Actions (CI/CD workflow)
• HashiCorp toolchain (Terraform, Vault, Consul)
• AWS (ECR, ECS-Fargate, Lambda, API Gateway, DynamoDB, Athena/Redshift)
• Airflow on Amazon MWAA (AthenaOperator, PythonOperator, EcsOperator), etc.

Bussiness Unit:

• ARV core

Responsibilities include:

• Develop in-house ML technology in a Computer Vision field
• Develop Object Detection models detecting crack defects on flare stack at offshore oil rigs
• Deploy and maintain Machine Learning API services, perform experimental MLOps technologies on existing ML services
• Support ARV’s subsidiary companies with AI and Machine Learning solutions
• Develop the company’s AI Studio Platform (initiation stage)

ML models and Application tool stack:

• ML framework: PyTorch, YOLOv5
• Application framework: FastAPI/Django, REST API, Docker, Celery with RabbitMQ

Application development tool stack:

• FastAPI/Flask, Celery, Redis, Docker, etc.

Deployment tool stack (work cooperatively with DevOps):

• GitLab Runner (CI/CD workflows)
• AWS (API Gatway, Lambda, Application Load Balancer, ECS-Fargate, ECR)
• Database tools: Object relational mapping tool SQLAlchemy, PostgreSQL, etc.

Bussiness Unit:

• 先端生産技術研究室 (Advanced Production Technology Research Section)

Responsibilities include:

• Develop machine learning models to improve materials manufacturing process
• Develop a web application platform to deploy ML models (using Django, REST API, jQuery)
• Optimize manufacturing process using machine learning techniques improving process efficiency, improve KPI, reduce cost, and prevent failure)

ML model methodologies:

• Logistic Regression, Multiple Regression
• Multi-Layer Perceptron Neural Networks, Convolutional neural network (CNN)
• Grad-CAM class activation visualization, RNN/LSTM, Autoencoder
• Bayesian Optimization, Gaussian Process, Markov Chain Monte Carlo (MCMC)

Data science basic knowledge:

• General statistic, Central limit theorem
• Maximum likelihood/cross-entropy optimization for various type of classification models
• K-fold cross-validation, Early Stopping Method, Model Regularization and Generalization, Overfitting Prevention

ML frameworks:

• Keras-TensorFlow, Chainer
• SciPy, Numpy, Pandas, Scikit-Learn, Scikit Optimize, Hyperopt

Deployment frameworks:

• Django, REST API, Basic Vanilla JavaScript, jQuery, Chart.JS
• Apache Web Server, Private Internal Windows Server

My Work Achievements:

• Successfully developed a Python-based Data Analytics Web Application for internal use of production engineers using the Django framework
• The web application features include the utilization of machine learning models for prediction, visualization, and optimization
• The application also featured an basic interactive GUI using jQuery and basic HTML template so that production engineers can use the web application features to optimize manufacturing parameters such as the chemical composition of an alloying element, the reduction rate of a cold rolling process, heat treatment time, etc.