Designing and building out the ML components of BCG's Fabriq personalized marketing platform

Designed the ML components of Formation's Offer Optimization Platform. Built out Formation's general-purpose AWS cloud-based MLOps system for rapid model development, deployment, workflow orchestration, and monitoring, including core infra, CI/CD, workflow management, and monitoring.Developed and deployed multiple machine learning models (behavior propensity, product recommendation) using PySpark, SparkML and AWS Sagemaker to 1:1 personalize millions of marketing offers on behalf of Formation's large enterprise clients. Implemented a multi-armed bandit / contextual bandit reinforcement learning system for customer-level context-based offer selection and continuous campaign optimization. Responsible for building out data pipelines and pipeline orchestration via Kinesis, Glue, Spark, SparkStreaming, Airflow, Lambda, and Step Functions. Experimental design and analysis (power analysis, stratified sampling, matching, covariate-adjustment) to rigorously measure campaign ROI/Lift wrt held-out control groups. Back-end service design and implementation for data products in Golang/GraphQL.

Designed and prototyped a SaaS application to optimize production throughput for a large aluminum manufacturer using combinatorial optimization and mixed integer programming. Estimated lift in throughput of roughly 25% during pilot phase testing and implementation. Contributed to an online streaming anomaly detection and classification system for large multinational steel producer. Processed terabytes of IoT production line machine sensor data using signal processing techniques and multivariate anomaly detection models with Apache Spark. Conducted pattern matching and time series analysis to map new streaming anomalies to existing pattern libraries and correlate anomaly occurrences with machine failures for real-time alerting. Designed and implemented a market segmentation and dynamic pricing application for a major US airline. Used clustering methods to automatically segment market demand for hundreds of long-tail origin-destination pairs and automated pricing experiments to continuously adjust fares in response to market-specific elasticity. Estimated revenue lift of 5% in A/B testing over 60 pilot markets.

Product analytics in support of Uber's Vehicle Solutions program. Data analysis to understand user behavior, identify key drivers of growth/churn, drive feature recommendations, and gauge effectiveness. Experimental design & analysis (A/B Testing) of new product features. ETL pipeline development in Apache Hive and dashboard development in RShiny.

Developed several credit risk predictive models for real-time loan application decisioning and adaptive risk-based pricing. Built R and Python packages to streamline the model building and evaluation process, and automatically gather publicly available data for business loan applicants to enrich the automated underwriting process.

Conducted econometric modeling and causal inference analysis in support of multiple public policy evaluations and technical assistance contracts for government (federal, state, local) agencies and private foundations. Complex experimental and quasi-experimental design and analysis with both randomized and observational data. Authored white papers including reports to Congress, and delivered presentations to clients and at academic conferences. Published research findings in academic journals.