Building ML systems for media and catalog understanding

• Developed a Machine Learning platform for forecasting of labor and sales data through an on-prem to GCP migration.

• Managed and optimized data pipelines to support 10s of terabytes of data to create weekly time-series forecasts for 100 million product-location combinations.

• Led MLOps to support demand forecasting including spark jobs optimization, integration testing, and CICD.

Tech Stack: Hadoop ecosystem (Spark, HDFS, YARN), Workflow management (Airflow), Distributed SQL query engines (Hive, Trino), GCP services (BigQuery, Dataproc, Cloud Composer, GCS), Python

• End-to-end design and implementation of microservices for AWS customer demand forecasting.

• Developed dashboard displaying AWS service usage metrics, utilizing Vue.js for the front-end and Scala for the back-end. Implemented CRUD operations interfacing with metrics stored in Aurora RDBMS.

• Event-driven orchestration of a highly manual forecast generation process allowing the team to increase the cadence of forecast generation from annual to weekly.

• Master’s thesis on Behavior Detectors to Support Feedback Generation advised by Prof. Neil Heffernan

• Research work aimed at helping reduce workload on teachers and allowing them to spend more time providing individual attention to students.

AWS software capacity and revenue forecasting: building infrastructure to run predictive models of AWS service usage and store/display the results in an accessible format for AWS service owners

• Created dashboard to display AWS service usage metrics. Webpage written in vue.js and a scala backend. Backend interfaced with an Aurora instance and provided CRUD operations to interface with metrics

• System design and implementation to extend existing model running infrastructure to implement the backtesting of these models in order to scale up to 100s of concurrent backtests previously run individually on a local machine