1) Built customer service chatbot for small and micro enterprises loaning; participate in building the NLU module of the KBQA, and task chatbot. Collected and preprocessed corpus marked by the outsourcing personnel according to the needs of the business scenario, and expanded the knowledge base; Embedding using language model Word2Vec, FastText, StarSpace, Bert, SentenceBert,RoBERTa (Based on Transformer architecture and Attention mechanism) etc. Downstream tasks including textCNN, naiveBayes, BiLSTM, Bert for classification, andcalculating cosine similarity for text matching. Deployed solutions as inference APIs/pipelines and used the stratege of retrieval and re-ranking.Our chatbot helps the company promote sales as well as reduce costs of Telesales staff by 80%.2) Zero to One, from the bottom up, take charge of the prepayment forecast project of the dynamic pricing strategy of loan products. Based on 200+PBOC(People’s Bank of China) credit data, customers’ application data etc stored in the big data platform(Hadoop), and the data volume is tens of millions of records. Performed sampling, features engineering, variable screening, model training, inference and A/B tests; built robust models using XGBoost, Random Forest, Logistic Regression. The KPI of prepayment prediction was increased by 15%.3) Take charge of digital marketing project to whitelist potential customers. Based on Ping An Group’s 3000+ features(personal info, AUM info etc) and outbound callingdata of Shanghai remote sales center. Performed sampling and statistical analysis; eliminated noises; dealt with sensitive data;provided user portrait of specific loan products, and whitelist of potential customers.4) Responsible for providing performance management guidance to telesales staff, doing sampling and K-Means cluster analysis. My sales model and strategy helped the company increased loan amount per 100 name lists.

1) To solve the problem of classification of online products without bar code, participated in text classification of 178 categories of commodities from tens of millions of records every month. Preprocessed text data comes from major e-commerce platforms including Tmall Taobao(Alibaba Group), JD.com, Suning.com, etc.2) Exploited algorithms such as SVM, FastText; model precision& recall was more than 90% on test sample set; inference results were generated into Oracle database. Since previous manual way was replaced, efficiency was improved and more than $100,000 costs was saved for the company.3) BAU monthly, evaluate the new product definition, produce analysis reports, and do anomalydetection to ensure the quality of the generated data.

1) Responsible for 10+ projects using Agile software development strategy in a team of 8 members. provide decision support.2) Using Informatica to ETL(extract-Load-Transform) data from different databases (such as SIEBEL) every day automatically.3) Data visualization using SAP datashboard; wrote SQL VIEW, SQL material view, the designed real-time service request monitoring dashboard is displayed on the 3mX2.7m large screen of the COMAC quick response center.