ML new initiatives and whole-page optimization.

As a founding ML engineer on the team, I built ML from scratch and currently lead all user-related algorithms: home screen content and category ranking, onboarding ranking, autoplay, linear ranking, internationalization ranking, related title, image ranking, personalized email/push notifications, user acquisition, etc.I built the first A/B test framework, including exposure event tracking, metrics calculation, significance calculation and correction, data pipeline, dashboard, analysis tools, and best practices across the whole company. This increased A/B tests from 0 to 100+ per year for the ML team.I was deeply involved in the first generation of ML and data infrastructure design and implementation: model serving, vector database, feature store, offline metrics calculation, A/B test integration, analytics events, etc.

I create new Data Science applications, deliver end-to-end solutions, and spread the fun of Data Science here at Trulia: - Computer Vision Applications- Natural Language Understanding / Text Search- Natural Language Generation- Real-Time API Infrastructure- Web Demo for Every Project- Organize Data Science Lunches at Trulia- Mentor Capstone Projects in CMU-MCDSTech Stacks:- API Infrastructure & Web Development: Spring Boot, Django, Flask, Solr, Redis, Docker, ELK (Elasticsearch, Logstash, Kibana)- Data Analysis & Pipeline: Hadoop, Spark, Pandas, Oozie

AI and Law, OAQA Lab, School of Computer Science, Carnegie Mellon UniversityLUIMA ("Legal" UIMA) is a UIMA-based natural language information extraction framework and toolkit tailored for the legal domain. It aims at providing base functionality in processing legal text (cases, statutes, etc.) that can be used to enhance or jumpstart development of systems in legal information retrieval, text classification, semantic extraction or similar applications.My research work:- Use Machine Learning and NLP tools to annotate documents based on UIMA- Index the annotations with Solr/Lucene- Use Learn to Rank modules to improve the search resultsThe results were published in the Proceedings of the 15th International Conference on Artificial Intelligence and Law: http://dl.acm.org/citation.cfm?id=2746096

Business Process Management Lab, School of Software, Tsinghua UniversityMy research work:- Improve existing algorithms and designed new algorithms for process decomposition and reuse- Analyze the properties of process models from CMCC (China Mobile)- Results were published in Business Process Management 2013 Workshop: http://link.springer.com/chapter/10.1007/978-3-319-06257-0_32Developed a log analysis system for Panguso (盘古搜索), a search engine company (now known as ChinaSo):- Build the result display modules- Design and implement a new process-mining method adapted to company-specific dataI am familiar with process modeling languages such as Petri nets and BPMN 2.0 (Business Process Model and Notation).

Data Mining Lab, School of Software. Refined a topic model to address the problem of identifying topical experts in Question Answering Communities. Collected 10,000 questions and 50,000 answers from Yahoo! Answers. The results were published in the World Congress on Engineering 2012.

As a pioneering Android developer in Chinese universities since version 1.5, I've created innovative applications that have accumulated over 100,000 downloads.