華新麗華總部  •  AI 資料科學家

03/2023 - now

Digital Information Platform Development 

Project Management (PM): 

• Led the design and planning of management dashboards, overseeing everything from requirement gathering to business process optimization.
• Established enterprise data governance and metric standards.

System Analysis (SA/SD): 

• Designed and implemented an integrated lakehouse architecture, incorporating Databricks' bronze-silver-gold database framework. 
• Automated ETL architecture via M365 and Azure Data Factory, with data visualization in PowerBI.
  

Programming (PG):

• Developed Python APIs and Data Mart databases, integrating multiple data sources. 
• Implemented Python monitoring modules, using SPC quality management and Pandas, Scipy for statistical analysis to identify and address anomalous data.
• Control code quality w/ git and azure devOps for project versioning.

Core Project Experience

Internal Knowledge Management Bot: 

• Developed an Azure OpenAI GPT-4-based generative chatbot (RAG) for technical document retrieval and Q&A in the R&D department.
• Integrated health insurance databases and corporate regulations to develop a claims Q&A bot, utilizing Elastic Search for efficient document retrieval.

Stainless Steel Process Factor Analysis:

• Enhanced computational efficiency by 80% using Azure Databricks and Data Factory. Managed model lifecycle with MLflow, established CI/CD processes and conducted significant feature analysis for production yield optimization.

Stainless Steel Business Forecasting AI Prediction:

• Implemented Python web scrapers for economic data integration with MES and ERP systems for customer behavior analysis. 
• Applied NLP for CRM data modeling, using K-means and GBDT、XGboost for customer segmentation and sales prediction, achieving over 70% accuracy (baseline~30%).
• Pioneered the integration of PyTorch and HuggingFace's Large Language Model (LLM) to embed unstructured data from CRM systems. approach to enhance feature engineering and processes to optimize the regression model.

PowerBI Promotion Plan:

• Recognized as a top 10 instructor in 2023 for teaching dashboard information structuring and UI/UX design. 
• Provided cross-departmental dashboard system design consultations, aiding in the establishment of effective automated information flows and problem-solving solutions.

Big Data 巨量資料分析  •  資策會數位教育研究所

08/2022 - 01/2023

Automatically web crawler (Python, Selenium, BeautifulSoup, Regex, mySQL)

• Raw data extraction and overcome anti-crawler issues w/ 104.com, Medium, Ptt, Dcard.
• ETL for hybrid data among text, categorical and numerical type to establish the data warehouse.

Data-driven intelligence deliverables (Sklearn, Tensorflow, Pytorch, Hadoop, Spark)

• Familiar with EDA(Exploratory data analysis) in the semiconductor field based on strong statistical methods and data sense.
• Conduct the highly readable Tableau dashboards to visualize 600K rows of data to find out the potential influence factor.
• Ability to design important features and PCA technology to improve model complexity and performance, which decrease feature number 30%.

Machine learning (Regression, Decision Tree, PCA, Bagging and Boosting)

• Apply blending with Xgboost&NN in regression and achieve r square to 89%, RMSE to 6000+ for 104.com salary prediction.
• Explained Al model(XAl) by feature important and Shapley value to interpret the model behavior.
• Ability to Design simulation & parameter tuning experiments by Cross-Validation and Grid Search to optimize model performance.
• NLP tech applies b/w TF-IDF analysis and word2vec via optimized jieba/NLTK for text modeling.
• NLP tech w/ roBERTa model sentiment analysis based on 700K comments and partial hands labeling on work stage in PTT/Dcard which F1 score reach to 89.4%.

Deep Learning (ANN, CNN, RNN, LSTM, ensemble modeling)

• Experience in conducting experiments in various settings of datasets and model hyperparameters, and comparing the performance of experiment results with DNN and CNN model.
• Develop specific algorithms and neural network ensemble models for text data by tensorflow and Pytorch.
• Conv1D textCNN via skip-gram and CBOW W2V transform from 104.com work content and combined DNN model to predict salary which RMSE val-loss reach to 11000+.

Project

• 資訊業職缺 - AI推薦系統

CVD Process Engineer  •  Winbond Electron Corporation

05/2021 - 08/2022

AMAT GT/GT3 tool owner:

• Responsibilities for maintenance inline quality in DRAM 28nm tech node for IMD, PMD, PD, and HM layer, which mainly include SIO, SIN, SION filmAlso achieved PRD and collaborated with vendor site to complete acceptance test for KH factory in early stage.
• Experience in project management for NPW recycling, Originated DOE and optimized recipe structure to reduce overall NPW usage cost and reduce cycle time by 80%.