Image inpainting and Semi-Supervised Multimodal Learning.

Responsible for enhancing the Credit Risk Model, aka SoLo Score, to benefit over a million active users.

Developed and optimized data-driven solutions to enhance performance, scalability, and accessibility. This included productizing Data Export APIs with AWS Step Functions and Lambda for ad-hoc data retrieval, integrating AWS QuickSight dashboards to drive a 13% revenue increase, and leading a transition to a multi-layer DataLake on S3 and Athena, cutting maintenance costs by 60%. I also built a cost-efficient ETL pipeline with Spark and EMR, using functional programming for modularity, and created a dynamic ingestion module that reduced manual effort by 70% and improved reporting accuracy by 50%.

As the Founding Engineer, built ETL pipelines orchestrated using Airflow, alongside the development of campaign customization and tracking platform with Django and React. Also, designed an A/B Testing module for real-time experiments on customizable customer segments. Collaborated cross-functionally, to implement and deploy containerized RESTful APIs for data upload and recommendation delivery on Kubernetes. Additionally, architected robust data ingestion APIs on AWS for seamless data integration from heterogeneous data sources.

Enhanced the personalization model by transitioning from item-item collaborative filtering to integrating contextual features, incorporating user preferences, and assigning variable importance through a neural network-based architecture. Also, introduced a contextual multi-armed bandit approach to effectively address real-time user preferences and contextual factors, improving content delivery optimization. By incorporating impression discounting and dithering techniques, increased user exploration and the discovery of new products. Moreover, successfully deployed recommendation models on a Kubernetes cluster using Tensorflow Serving with versioning for seamless management.