• Processed data by extracting time and space features, and separated by different brands and parts, also collaborated with Johnson & Johnson to address data scarcity by successfully validating assumptions.• Reduced surplus inventory and enhanced order fulfillment efficiency through Linear Regression and CNN, analyzing spatial patterns in sales data and complex feature relationships.• Predicted future inventory of PCM by using LightGBM focusing on past sales units and sales brands, adding normalization andGridSearchCV, achieving an RMSE < 0.03, indicating an optimization of inventory management, boosting supply chain efficiency for orthopedic medical devices.

• Achieved the final 1800s operating curve, which is the first-time theoretical trial for Chinese road situation, through classifying the operating curve slices with K-means, constructing alternative curves by time ratio, and using CRITIC method to empower features.• Performed sentiment analysis by implementing Transformer architecture, specifically BERT, for natural language processing tasks, achieving a 15% improvement in accuracy.• Constructed a strategic map by matching travel data of vehicles with the reality, and combination of OD data, the layout of NIO’s overcharge pile and travel heat with Folium, covering more than 100 thousand users, providing data support for the next generation of new car users.• Ticket PLUS: Java Web Service – Event Search Engine and Ticket Recommendation: Front End:• Developed an interactive web page for users to search events and purchase tickets (HTML, CSS and JavaScript, AJAX)• Improved personalized business recommendation based on search history and favorite records. Back End:• Created Java servlets with RESTful APIs to handle HTTP requests and responses.• Built relational and NoSQL databases (MySQL, MongoDB) to capture real business data from Ticketmaster API.• Designed algorithms (e.g., content-based recommendation) to implement business recommendation.• Deployed server side to Amazon EC2 to handle 150 queries per second tested by Apache JMeter.