Taiwan
Results-driven AI leader with 10 years in data science, applied machine learning, deep learning, and LLM applications — leading a lean, high-impact team driving digital transformation through end-to-end AI pipelines and production-scale infrastructure.
• Founded and led Shopee Taiwan’s first Data Science team, a lean 4-member team deploying 40+ production models (33 active by end-2025) and 10+ automation pipelines across platform-wide risk and operations, scaling organizational capacity by 58 FTEs (~NT$45M annual efficiency gains). • Architected a multi-step LLM reasoning pipeline combining summarization and risk-type classification; automated 450K+ unstructured CS logs monthly with 95% automation, streamlining triage equivalent to 9 FTEs. • Owned delivery of a RAG-based LLM system for off-platform directing detection; successfully auto-passed 1.2M dialogues (80% automation), unlocking operational capacity equivalent to 12 FTEs while driving an 11x increase in confirmed detections with a near 0% appeal rate. • Drove a multimodal BSMI infringement detection system by fine-tuning LLMs (QLoRA/Unsloth) to verify product safety certifications; achieved 96% precision, clearing a 600K backlog and automating 80% of new daily arrivals. • Led the development of an SBERT-based semantic matching system for NCC regulatory compliance; automated 358K annual audits with 99.3% precision and a 0.02% appeal rate, reducing manual labor costs by 53% • Spearheaded scalable fraud detection models using LightGBM across new registration, existing, and buyer-stage users; automated 8M+ reviews and froze 1.4M+ high-risk accounts in 2025. • Initiated a unified defense mechanism with AML, Fraud, and Payment teams, integrating detection pipelines that identified 10K+ associated accounts, 400+ money-laundering cases, and 2K+ frozen rented accounts. • Developed 5 specialized NLP models using Bi-LSTM for regulatory and compliance item detection, achieving 97% precision and absorbing workloads equivalent to 3.4K review days annually (~14 FTEs).
• Apply polynomial regression model to build sales estimator system which estimates 400+ competitors' product daily sales. • Use LSTM, POS and, NLP skills to develop Review Sentiment Classifier for product review classification and clustering at sentence level, accelerating key sentence extraction time from eight hours to 30 seconds, saving 99.9 percent of time. • Develop Text Generator by Language modeling skill, generated 10,000 reviews for training purpose, which improved 2% accuracy(from 94% to 96%) in Review Sentiment Classifier. • Lead and import data-driven business transformation, from data collection pipelines, data structure definition to automated analysis dashboard, improved efficiency by reducing 90% of the time required for product managers and designers to extract useful insights. • Construct Captcha breaker by object detection and image recognition techniques to assist data engineers achieve auto crawler. • Conduct social network analysis to extract potential influencer and explore association in social network, saving 80 percent of time for reviewing useful feedback. • Deliver statistical data analysis and visualization tools in data service, including user behavior, product sales performance, clustering (k-means), association rule and market segment, which helps traditional retailers find useful insight and business strategies by data-driven analysis. • Devise and create B2B marketing performance data visualization report by Data Studio; provide useful insight and analyze existing problems in SEM/website performance weekly review, report directly to assist business strategy decision making.
• Crafted Fault Detection and Classification chart detector system, saving 97 percent FDC U-chart within spec abnormal inspection time. Utilized Spotfire to show data visualization and design an interactive Graphical User Interface for data analysis result. • Employed k-means method in “U-chart by part analysis”, grouping similar parts together and tightening existing control spec.
System maintain and trouble shooting Process improvement Project management Data mining and Statistical data analysis
• Used 5S method to reduce 50% working time in receiving dock. • Save $20,000 USD inventory cost through warehouse management. • Created a new working station in an urgent new project.