Data Scientist Intern • AU OptronicsJuly 2022 - Sept 2022 | Research Assistant • NCKU CSIEMar 2022 - June 2022 |
• Achieved 95% accuracy in machine defect detection by training YOLOv5 and 93% on the SOTA YOLOv7 model. • Implement alarm system to notify engineers when defects are detected with Python. • Designed and developed user-friendly UI using PyQt5 to streamline label generation and model retraining, empowering non-technical users to accomplish these task with ease and efficiency. | • Utilized Python and various data analysis tools (NumPy, Pandas, Matplotlib) to implement standardized approach to bounding box labels, contributing to improved model performance. • Leveraged YOLOv5 with accuracy of up to 90% accuracy, consistently delivering reliable results. • Develop dataset auto-labeling pipeline using Python and trained YOLOv5 to increase labeled dataset. |
Sept 2022 - Aug 2024
• Focus on SLAM & monocular 3D scene reconstruction research to improve model performance and/or reduce computational cost and develop AR application system pipeline with Unity for tele-meeting scenario.
• Related coursework: Intro to Fintech, Game Programming, Robotics, Artificial Intelligence, Advanced Computer Vision
• Average GPA: 4.0/4.3
Sept 2018 - July 2022
• NCKU x Bank Sinopac Scholarship and Encouragement Student Aid (Jan. 2021)
• Related coursework: Data Structure, Algorithms, Operating Systems, Computer Vision, Deep Learning
• Average GPA: 3.0 / 4.3
• Developed an automated Python trading system with FastAPI as the backend.
• Analyzed daily profit/loss data and automated comparisons with NASDAQ.
• Employed MongoDB to store profit/loss data efficiently.
• Ongoing work: Developed a web interface to visualize profit/loss charts using HTML, CSS, and JavaScript (Chart.js).
• Ongoing work: Exploring machine learning models for stock price predictions
• Leveraged the cutting-edge Langchain framework to interface with GPT-3.5-turbo LLM API.
• Designed a user input pipeline system, capable of generating extensive 1500-word blog content.
• Improved pipeline efficiency by 40%, reducing inference time from 5 to 3 minutes.
• Enhance content accuracy and reduce LLM hallucination with Google SERP data scraping.
• Employed Flask framework for efficient API communication.
• Applied prompt engineering to optimize LLM output.
• Designed a weight tracking website using HTML, CSS, and JavaScript.
• Utilized Chart.js to create visually stunning weight charts.
• Developed a Node.js backend for data transmission.
• Stored and managed weight data efficiently in MySQL.
• Developed a mobile app for image capture and content browsing using Flutter.
• Achieved 93% accuracy in traditional Chinese medicine detection using YOLOv5.
• Employed FastAPI to interact with a custom-trained YOLOv5 model hosted on a server.
• Employed data analysis tools (NumPy, Pandas, Matplotlib) to analyze data seasonality.
• Utilized ADF and KPSS tests to assess data stationarity.
• Conducted model comparison to identify the most effective machine learning approach.
• Captured 1200+ images across 5 garbage classes.
• Achieved 92% accuracy in garbage classification with MobileNetV2.
• Achieved 95% accuracy in garbage classification with custom ResNet50.
• Created a Flask-based backend server for image inferencing.
Data Scientist Intern • AU OptronicsJuly 2022 - Sept 2022 | Research Assistant • NCKU CSIEMar 2022 - June 2022 |
• Achieved 95% accuracy in machine defect detection by training YOLOv5 and 93% on the SOTA YOLOv7 model. • Implement alarm system to notify engineers when defects are detected with Python. • Designed and developed user-friendly UI using PyQt5 to streamline label generation and model retraining, empowering non-technical users to accomplish these task with ease and efficiency. | • Utilized Python and various data analysis tools (NumPy, Pandas, Matplotlib) to implement standardized approach to bounding box labels, contributing to improved model performance. • Leveraged YOLOv5 with accuracy of up to 90% accuracy, consistently delivering reliable results. • Develop dataset auto-labeling pipeline using Python and trained YOLOv5 to increase labeled dataset. |
Sept 2022 - Aug 2024
• Focus on SLAM & monocular 3D scene reconstruction research to improve model performance and/or reduce computational cost and develop AR application system pipeline with Unity for tele-meeting scenario.
• Related coursework: Intro to Fintech, Game Programming, Robotics, Artificial Intelligence, Advanced Computer Vision
• Average GPA: 4.0/4.3
Sept 2018 - July 2022
• NCKU x Bank Sinopac Scholarship and Encouragement Student Aid (Jan. 2021)
• Related coursework: Data Structure, Algorithms, Operating Systems, Computer Vision, Deep Learning
• Average GPA: 3.0 / 4.3
• Developed an automated Python trading system with FastAPI as the backend.
• Analyzed daily profit/loss data and automated comparisons with NASDAQ.
• Employed MongoDB to store profit/loss data efficiently.
• Ongoing work: Developed a web interface to visualize profit/loss charts using HTML, CSS, and JavaScript (Chart.js).
• Ongoing work: Exploring machine learning models for stock price predictions
• Leveraged the cutting-edge Langchain framework to interface with GPT-3.5-turbo LLM API.
• Designed a user input pipeline system, capable of generating extensive 1500-word blog content.
• Improved pipeline efficiency by 40%, reducing inference time from 5 to 3 minutes.
• Enhance content accuracy and reduce LLM hallucination with Google SERP data scraping.
• Employed Flask framework for efficient API communication.
• Applied prompt engineering to optimize LLM output.
• Designed a weight tracking website using HTML, CSS, and JavaScript.
• Utilized Chart.js to create visually stunning weight charts.
• Developed a Node.js backend for data transmission.
• Stored and managed weight data efficiently in MySQL.
• Developed a mobile app for image capture and content browsing using Flutter.
• Achieved 93% accuracy in traditional Chinese medicine detection using YOLOv5.
• Employed FastAPI to interact with a custom-trained YOLOv5 model hosted on a server.
• Employed data analysis tools (NumPy, Pandas, Matplotlib) to analyze data seasonality.
• Utilized ADF and KPSS tests to assess data stationarity.
• Conducted model comparison to identify the most effective machine learning approach.
• Captured 1200+ images across 5 garbage classes.
• Achieved 92% accuracy in garbage classification with MobileNetV2.
• Achieved 95% accuracy in garbage classification with custom ResNet50.
• Created a Flask-based backend server for image inferencing.