Recommendation Systems on Destination----------------------------------------------------• Spotted shortcomings on existing recommender and proposed learning-to-rank algorithm. Led Destination Recommender effort from idea initialization, modeling, scale training, pushing to production, and A/B test implementation. Increased click through rate by 43% on homepage and 112% on app home feed.• Engineered and executed 3 end-to-end ETL pipelines within big data environment to provide essential training data at scale for recommendation models using PySpark. Built offline evaluation pipeline to compare models using ndcg, precision@k, mrr etc...Recommendation Systems on Listing-----------------------------------------------------• Developed personalized listing recommender with LightFM, overcame cold start problem by incorporating listing and user features.Built several listing recommendation models to production, including matrix factorization based on ALS, Glove and LDA.• Built Multi-Arm-Bandit infrastructure, enabled automatic daily traffic re-allocation and online evaluation on several recommendation models. The MAB system boosted efficiency significantly compared to traditional A/B tests.Natural Language Processing on Call Transcripts and Reviews--------------------------------• Identified and summarized call topics for Customer Experience department by Guided-LDA topic modeling. Built classifier to assign call transcript to topics in production, enabled real-time trending topic monitoring for management by ElasticSearch.• Applied Xgboost to enable automatic moderation to approve or decline listing reviews on highly imbalanced training data. Optimized the working flow for reviews team.

Customer Sentiment Analysis• Sentiment analysis and topic modeling on twitter feeds to track and evaluate performance of FedEx and its competitors.

E-commerce Opportunities Targeting• As analytical project lead, explored data sources and designed business rules, parsed millions of unstructured websites for relative regular expressions, performed dimension reduction and built predictive models to target e-Commerce customers for sales to pursue, built monthly repeatable process• Example: The count of keywords “shopping cart/my cart” on website homepage is the most significant variable in all models that helps identify e-commerce companies. Backwards selection selected 14 out of 33 keywords counted• Outcome: Generated 33K pristine e-commerce customers leads, designed web portal for sales to select and process leads, resulting in 10K incremental ADV (average daily shipping volume) and $11M incremental revenue• Tools used: Python (packages: pandas, bs4, scikit-learn, numpy, datetime, cx_Oracle, re), SAS, TableauEbay/Amazon Marketplace Analysis• Conducted risk assessment, performed statistical analysis on large datasets, looked for customers that are using Ebay/Amazon discounting programs, generated new insights and guidance for marketing strategies• Tools used: Hive, SQL, SAS

 Pricing Analysis on High Pricing Power Customers • Identified customers that are priced lower/higher than comparable companies. Defined customers with predictive modeling, segmented customers based on industry/product affinity, developed website to show analytical metrics of price/volume adjustment• Example: The largest segment (out of 7 segments) customers ship Ground Deferred and are in goods industry mostly with low yield• Tools used: SAS, SAS E-miner, Oracle, CGI, HTML, CSS, Javascript, Jquery, UNIX

Analyzed Billing Systems, constructed Billing Audit Report using SQL and SSRS to identify billing errors such as discrepancies and duplicates, looked into the root causes and corrected them by creating credit reports, largely improved accuracy in both automatically and manually billing system. Managed conflict situations occurred in customer charge calculations, produced analytical view, generated quantitative metrics and excel pivot table / chart reports adding VBA objects. Maintained databases, evaluated warehouse department performance by deducting credits according to incidents. Designed screens and forms in VBA system to support managers’ business decisions, tested and debugged systems