My team is at the core of the construction of the AI-powered Safeway Sincerely Health App. With a state-of-the-art Graph DB at the back-end holding the Customer-Nutrition Knowledge Graph, this app will become a customer's Personal Health Assistant, leveraging the domain expertise of nutritional experts and the AI expertise of our data scientists, using methods and tools such as Reinforcement Learning and LLM. It will offer exceptional nutritional insights and recommendations to our customers that will most likely lead to better health outcomes.

Helping to develop the next generation of exceptional data science leaders

I ran a world-class data science team in Capital One. We use the latest framework and tools to build highly predictive ML/AI models to identify each customer for a smooth sign-up and sign-on experience and to counter ill-intended behavior. In addition, we are combining the aforementioned with graph modeling (Connected Components, PageRank, etc.) to create an enterprise-wide platform for querying and traversing various internal System of Records (SORs) to associate similar and related customers. It's an exciting and under-explored area. We're trailblazers by definition.

I lead a high caliber team of data scientists and data engineers to tackle some of the toughest problems across the enterprise using AI

We (Zenalytics) are an internal venture (startup) of Bluestem Brands, a private retail/lending company with ~2b in annual revenue. In the past 3 years, I led the team in achieving the following:- Built from scratch multiple iterations of ETL and ML scoring pipeline and platform. Performance is comparable to Spark with similar number of CPU cores and similar amount of memory- Developed methodology and built adverse-actionable (AA-able) ML scorecard model for determining the credit-worthiness of credit applicants, demonstrated to be vastly superior to big bureau scorecards in enterprise testing, currently in production- Lead team in designing and deploying personalized add-on recommendation system for ourretail customers, leading to a substantially better customer experience with fewer but betterfitting offers while maintaining revenue- Lead team in developing and deploying real-time holiday sales forecast models, enabling theenterprise to optimize resources deployment during the peak sales period- Built and deployed machine learning models for all stages of a customer’s lifecycle: targetedmarketing, credit approval, first payment default, credit-line adjustment, near/long-termintention to pay, charge-off, collections- Used Deep Learning tools such as H2O, Theano, Tensorflow and Keras for building credit decisioning models and benchmarked the performance gain (in terms of model-fitting time) using GPUs- Built and deployed ML models using the Domino data science tools

At Azure Cloud:- Lead team in automating sentiment analysis from social signals (Twitter etc.) for quantifying the impact of favorable and unfavorable events related to our business; built dashboards and Word2Vec-based models.At MSN and MSN Apps:- I was an architect and a key contributors for building out end to end a brand new telemetry infrastructure from scratch for the new MSN and MSN apps across all major platforms (iOS, Android, Windows). I helped introduce a framework from a statistical point of view and made sure the system is designed for fulfilling analysis needs instead of being just a piece of software engineering. I generalized user action on different devices and came up with clear descriptions of various elements on an UI to be captured.- I lead team in building out real-time reporting using Omniture for the new MSN and MSN Apps. In the process I have gained tremendous understanding in the dimensions and segments needed in the tool to identify opportunities for growing our business.- To ensure the data that we are getting from the new MSN and MSN Apps are trustworthy, I have independently built a fully-automated pipeline for monitoring and alerting anomalies in the data.- I have spec'd, project-managed and tested a separate video telemetry and reporting pipeline for the new MSN and MSN Apps to capture users' watching behavior in order to drive better engagement and more accurate targeting for advertising.- Before the new video telemetry and reporting was built, I had been the maintainer of the legacy system that submits scripts for MapReduce processing, imports processed data into SQL DB and generates cubes and dashboards. I have learned the ins and outs of such a system and applied my learning to the design and implementation of the new system.

At Bing Ads R&D:- I have built ML models to model positional factors on the new Bing UI, accounting for the effect of annotations. The extracted features (predictors) were then used in fitting logistic regression models in the MapReduce cluster. The outcome of the study helped greenlight the change that started Bing's steady climb in market share since 2013.- I lead team in syndicating the most comprehensive user engagement impact study ever in Bing Ads by comparing the distribution of ad quality across all keyword segments. It helped bring focus to addressable issues instead of being constantly distracted by noise. It also shifted the analysis paradigm from per-ad to per-user basis.- To identify gaps in ads quality among leading search engines, I have conducted a competitve ad relevance analysis to find out custom quality thresholds for different market segments, using human judgment data fitted with Random Forest and SVM. It lead to the implementation of adaptive thresholding features that helped significantly improve the whole-page relevance across all segments, while holding RPS constant or better.At Search Business Group:- I lead team in conducting advertiser demand analysis via large-scale auction simulations and identified categories of industries to drive for sales for the highest ROI. The conclusions lead to a focused approach for the field sales team that resulted in an RPM bump.- In order to achieve win-win with our publishing partners, I have carried out threshold tuning and optimization for throttling qualified ad inventory that helped achieve a great balance between revenue and relevance. - I have made use of the methodology developed in my doctorate dissertation to model and understand the effectiveness of the four major marketing initiatives targeting premium advertisers at the time. It resulted in the identification of the only effective initiative and helped save money and resources worth hundreds of thousands of dollars.

- Assisted in designing course material, providing team leadership, supervising 6~8 other TAs.

- Assisted clients from on and off campus for problems in all statistical areas.

- Performed data quality and risk assessment, meta analysis on drug safety data on 10+ data sets, involving multiple high-profile drugs- Applied innovative graphical methods for pattern identification.

- Built predictive models, performed risk analysis of health plan customers.- Applied machine learning techniques to 7 million customer visiting records.

- Taught Applied Statistics course for social science and biology majors.- Independently developed lecture notes, homeworks, quizzes, exams; supervised one TA.

- Assisted technology development of next generation high performance Atomic Force Microscopes.- Performed pilot process testing and diagnostics, yield improvement data analysis.- Assisted design of capacitive sensor for dielectric characterization, experimental data analysis.- Modeled tip-sample interaction under capillary force in tapping-mode AFM.

- Designed and fabricated a novel AC electrokinetic BioMEMS micropump.- Performed digital image processing (based on maximum likelihood) for the study of microflows.