• Developed an EDM recommendation ensemble model (Bias SVD + XGBoost), resulting in a 10% improvement in email open rates. • Built KPI and operation dashboards, leading to recognition as the 2022 Best Employee. • Designed an ink forecast model to help product managers optimize ink procurement. • Utilized a collage model to assemble new images from company picture materials based on descriptive input. • ETL printing volume data from AWS S3 and integrated Teams to monitor the data pipeline. • Identifying the print trends and tagging the values of customers gave sales useful insights.

• Utilized XGBoost to assess the value of offline scooters and generated battery swap priority. • Developed an LSTM model to forecast rental demand based on user app events, enabling a better understanding of the fulfillment rate in each operational region. • Evaluated the value of parking lots, deprecated the unused for cost down, and identified potential places for better user experience. • Collaborated with colleagues, building a data warehouse, and data mart. • Applied Airflow to ensure the stability and accuracy of data. • Cooperated with the operation team and developed a scooter idle time dashboard which effectively improved the service level.

• Crawled articles from various websites and built an NLP ensemble model: Text-CNN, Logistic Regression, and LightGBM, achieving an accuracy of around 91%. • Analyzed article content to calculate content similarity, then recommended related articles to readers, which improved click rates and page views. • Integrated Google Analytics data, exported it to BigQuery and visualized it using Data Studio to help company partners grasp website traffic. • Advertising optimization, analyzed the performance of different kinds of advertisements to the business department to sell ad slots.

• Developed user segmentation to build CRM helping client administering their customers. • Implemented Xgboost to find the most possible purchase customer list and used Cosine Similarity to recommend their interesting product category causing the repurchase rate 4% improved. • Built exchange rate API compared the current exchange rate with past 30 days then demonstrated exchange recommend rank to users.