Jabil Corp. - Data Scientist, 2019 Sep - Now

Jabil (NYSE: JBL) is a manufacturing solutions provider that delivers comprehensive design, manufacturing, supply chain and product management services. Leveraging the power of over 200,000 people across 100 sites strategically located around the world.

Responsibilities : 

 1. Using modern deep learning tools to perform cosmetic defect detection.

 2. Data visualization, make data speak to help factory to make decisions.

 3. Leading-in infrastructure concepts. 

 4. Design/Develop a infrastructure for scrum development. 

 5. Design/Develop automation control system with C#/C++. 

 6. Team work with external corporations in AI projects.

• Improve inspection efficiency by collecting data from the shopfloor. 
• Analyze data to help improve the system or inspect whether the demands from business side is doable. 
• Do researches for new ML models (recommender model, NLP model) or architectures which are suitable for the system. 
• Key Projects: 
• AI-based real-time Optical Inspection System 
• Objective: 
• Improve Inspection accuracy of AOI system by collecting data from manufacturing machine. 
• Reduce the requirement of intensive workforce for cosmetic inspection.
• Responsibilities: 
• Play as the key role in design and development the end-to-end AI-based solution. 
• Make sure the images/data are being collected from AOI is in correct format every single day. 
• Analysis 325 GB images/data from shopfloor every single day. 
• High resolution image (10,000-pixel by ~7,000-pixel) AI-based cosmetic inspection. 
• Cooperate with Microsoft data science team to brainstorm the better solution. 
• Build ML model to predict each product to see if there is any defect on the product. 
• Use Azure Machine Learning Workspace to train/evaluate the model.
• Achievements: 
• Reduce the overkill rate from 20% to 6.3%. 
• Build up the end-to-end deployment through Azure DevOps pipeline. 
• Deliver the AI-based solution to shopfloor.
• Tools: 
• Tensorflow (Keras): 
• Transform data into tensor object. - Build LeNet model. 
• Build Variational Auto-Encoder model. 
• Transfer learning from trained Efficient-Net. 
• Record metrics for trace the training process. 
• Save trained model weights for prediction. 
• Scikit-Learn: 
• Training set and testing set split. 
• Computing metrics (e.g., confusion matrix, ROC curve, AUC, recall…etc.)
• OpenCV: 
• Image preprocessing (e.g., find contours, fit rectangle, rotate, color masking…etc.) 
• Scikit-Image: 
• Find shifted x, y based on golden image. 
• Register translated images. 
• Image I/O: read image and save image. 
• Numpy: 
• Crop image with certain ROIs. 
• Numpy object mathematic batch computing. 
• Tensorboard: 
• Visualize the metrics from the whole training process. 
• Matplot/Plotly: 
• Data visualization (e.g., line chart, pie chart, scatters, confusion matrix…etc.). 
• Imgaug: 
• Enlarge data set by using image augmentation. 
• Pandas: 
• Manipulate data as table-liked object. 
• Read/write csv. 
• Docker: 
• Containerize .NET core micro services written by .NET C#. 
• Build docker compose to orchestrate micro services. 
• Azure DevOps: 
• Version control for codes by using Azure Git Repo. 
• Merge branches to master branch. 
• Working items (user story/task/bug) assignment, update by using Azure Board. 
• Design pipeline for auto-unit test and auto-deployment when there is any update of codes by Azure Pipeline. 
• Trigger auto-retrain pipeline for ML model when there is any update of ML-related codes. 
• Azure: 
• Data transfer from shopfloor by using Azure Data Factory. 
• Data storage by using Azure blob storage. 
• Train/evaluate ML model by using Azure Machine Learning Workspace. 
• RabbitMQ: 
• Message queue architecture to build asynchronous API call.
• Design topology of message queue (e.g., fan-out, multiple binding, subscribing…etc.)

Jabil Corp. - Programmer Analyst, 2017 Feb - 2019 Sep

Improve the efficiency of making products in the shopfloor.  

• Key Projects: 
•  Infrastructure (Common library) 
• Achievements: Boost the development time more than 10% faster. 
• Tools: .NET C# 
•  AGV (Automated guided vehicle) System 
• Achievements: Reduce the 10% workforce requirement in the shift. 
• Tools: .NET C#
• ◼ Automation System
• Achievements: Integrate the AGV system with automation system. 
• Tools: Code-gear C++
• ◼ MES (Manufacturing execution system) System 
• Achievements: Operator in the shopfloor can scan face to enter the MES system. 
• Tools: Code-gear C++

Innolux Corp. - Computer vision Intern, 2015 June - 2016 July

As a leading TFT-LCD panel supplier, Innolux has been deeply engaged in the research and development of TFT-LCD technology and production with its abundant innovative energy and insistence on high quality. Specialized in manufacturing TFT-LCD panel.  

• Key Projects: 
• Color tag categorization 
• Objective: 
• To distinguish different color of tags on the LCD panel. 
• Reduce the requirement of labor in the shopfloor to categorize the tag on the panel. 
• Responsibilities: 
• Develop a function integrate with AOI system to categorize the color on the LCD panel. 
• Achievements: 
• The function can 100% categorize different color tags. 
• Replace the human color-check station in the shopfloor by the system. 
• Tools: 
• .NET C#: Design the algorithm to identify the color of the tag.

AI Academy - Trainee, 2019 June - 2019 Sep

Experience : 

 1. Training AI technology/knowledge every Saturday. 

 2. Cutting-edge algorithm introduction. 

 3. Build up data science fundamental knowledge.

CSIE. - Southern Taiwan university of science and technology, 2012 Sep - 2016 June

Learning : 

 1. Operation systems 

 2. Data structure 

 3. Linear algebra 

 4. Discrete mathematics 

 5. Engineering mathematics 

 6. Software languages : C#/C++/Python/Java/Java script...etc.