Design Intelligence Laboratory (DILab)

+ White Space Project: Built a random forest model predicting ROI for each block group; identified 15K out of 220K block groups with high ROI, reducing the planning efforts of the real estate team by narrowing down the scope for investigation in North America+ Localized Pricing Engine: Migrated localized pricing solution from Databricks notebooks environment to Azure Cloud environment with Azure Data Factory, reducing run time by 88% (70 hours) and saving 76% cost ($500)+ Mentor Junior Scientists: Led and mentored junior data scientists to develop ML solutions for business stakeholders and scaling existing pricing solutions for convenience stores globally+ Localized Pricing: Improved a set of statistical models (time series modeling, regression) to recommend product prices, resulting in an impact of 3% average margin lifts ($120M+ gross margin) per year in North America+ Site Selection: Utilized a random forest regression model to predict revenue of potential stores, increasing prediction performance by 15%, avoiding real estate investment failure over $2M+ in Texas sector+ ETL Pipeline: Developed data pipelines to track store status, ensuring data quality for modeling and business purposes, saving 60+ hours of manual efforts each quarter

+ Conversions Analytics: Developed metrics to evaluate the current performance of attribution identifiers (e.g. IP address) with SQL; recommended adding phone numbers into attribution model, driving 40% increase in ads conversions+ COVID Dashboard: Created Tableau COVID Recovery Dashboards, tracking over 10,000K transactions of 10+ major clients for the past 2 years to decide the relaunch schedules of ads campaigns+ Report ETL Library: Created a Python library to automate data pipelines, ingesting data from various sources (e.g. Redshift, S3, FTP, Google Sheet, etc.) for daily reports sent out to 15+ clients+ ChatBot: Built a chatbot in Google Chat with Python and NLP packages (spacy, NLTK) to answer questions related to client campaign performances, and won first place in company’s Hackathon

+ R Library Development: Developed an R package with functions of data cleaning, cohort balancing using propensity scores, multiple hypothesis testing and data visualization for electronic health record data, which is used by 3+ researchers