Incorporated the latest advances in LegalTech, deep learning, and NLP for automatic legal invoices adjustment and alerting. Captured additional invoices missed by auditors on a 10 thousand scale.

• Developed, implemented and continuously improved machine learning models including CodeDetection model, ToxicSpeech model, ScreenshotDetection model, and Keywords model to empower the product's AI capability in social-collaboration platform including Microsoft Yammer, Teams, Workplace and Slack - Developed end-to-end neural network models using recurrent and "deep" architectures (developed in PyTorch) including data collection, model training, and model selection - Developed a topic model by latent Dirichlet allocation and matrix factorization to empower 'similar threads search' and 'top keywords' feature - Dockerized models on AWS as independent API endpoints to serve 10,000 requests per minute, per instance and built a model scoring pipeline by multithread programming to enhance performance• Designed and built a REST API to extract keywords from social messages for further social network analysis using the Flask framework• Enlarged training data sources by the back-translation technique in natural language processing in a relatively parsimonious way using Google Cloud Service• Illustrated the hidden relationships in customer networks by identifying the components and subgraphs using techniques including triple extraction and eigenvector centrality computation

• Collaborated with Institute’s Board of Directors and conducted feasibility analysis on establishing Fudan-CSG Foundation (worth 0.2 billion) for technology transformation on intelligent diagnosis- Conducted benchmarking analysis on operational metrics, times-series analysis on the financial statements of CSG and researched business model for key players in the intelligent diagnosis industry- Integrated data, designed and developed dashboards to showcase the insights on intelligent imaging diagnosis industry by pivot tables and Tableau• Increased the institute’s web traffic by 15% per month by analyzing the traffic data from MySQL database, identifying the poor performing pages by defining key metrics such as bounce rate• Improved the Big Five personality model with cross culture(America vs China) validity by Chinese Character segmentation, TF-IDF and TextRank algorithm using Python