- As a tech lead of MLOps team, led a group of data engineers and machine learning engineers to determine mechanisms for machine learning model lifecycle and release several sample codes such as unit test (Pytest) and MLflow which optimized the quality of online models.
. Created a Slackbot and a real-time scoring system on Google Cloud Platform to hold E.SUN AI Open
Competition for competing supervised learning and natural language processing techniques.
. Built an ML pipeline to extract blacklist of suspected financial criminals from the news and present the model results on a web frontend built with Vue.js which optimized the AML (Anti Money Laundering)
process. The system tripled the number of correctly caught criminals and reduced one-sixth of reading
articles compared to rule-based methods.
. Led 3 developers to launch a system predicting the probability that a customer would delay mortgage
repayment by LightGBM and SHAP This model has been online for one year, and it has not only helped the
card loan division improve repayment collection but reduced 68 person-days (56%) % per month.
. Led 2 developers to develop a predictive model and optimization algorithm to help customers find the right
financial assistants by LightGBM. This service helped the wealth management division activate customers
who had not traded for a long time and the transaction amount reached 4 million dollars during the
three-month trial operation.