Lions Gate Entertainment Corporation is an American, Canadian-domiciled entertainment & media company with well-known entertainment contents such as La La Land, Orange is the New Black and Wonder• Led a team of six to develop a thumbnail selection deep learning model to automatically select high quality thumbnail for Lionsgate and Starz videos to increase click rate and engagement of audience• Developed a new algorithm with instance segmentation, image style transfer and topology to transfer arbitrary style to a specific instance in arbitrary shape in an image, with potential use cases of image creation and editing in movie posters and marketing materials (paper mentioned below)• Built a pipeline with Python and AWS EC2 to daily gather videos and extract features from videos and audios using CNNs to augment existing video feature libraries, saving $36,000 annually to create dataset for future planned projects such as video search engine and piracy detection• Implemented and optimized Louvain Algorithm in Python to partition 250, 000 actors into different communities using AWS EC2 to facilitate and assist theatrical team in making casting decisions• Conducted churn analysis using Python for Lionsgate's Over-The-Top platform using time window sliding methodology to discover the importance of subscriber behaviors in a seven-day window

• Developed statistical geometric distribution model incorporating factors such as churn and subscription fee to estimate individual-level subscriber lifetime value on Lionsgate's online streaming platform• Built web scrapping program with BeautifulSoup in Python to extract user reviews of 300+ movies from the Internet that saved 5-10 hours and supplement data for movie popularity analysis• Performed rule-based sentiment analysis on user reviews of Lionsgate’s popular movies in Python and developed decision tree model that featured the connection between negative sentiment and movie popularity to drive movie meta acquisition

• Graded homework and exams of two MBA classes with over 100 students every two weeks• Provided feedback to instructor regarding common mistakes and overall results after each grading to facilitate in-class instruction