• Building (micro-batch) event ingestion system for our ML platform which materializes features from interaction data and publishes to online and offline feature stores
• Engineered A/B testing platform to power online experimentation for recommendations rails
• Refactored data pipeline code to be configuration driven, allowing for logic to be ported to international regions

• Kinesis
• DynamoDB
• Databricks Delta Lake
• Scala
• Python

• Built online embedding-based retrieval (EBR) and reranking platform with Elasticsearch to serve recs in real-time:
1. The system serves all homepage traffic, handling ~50rps with p95 latency <60ms
2. This work was part of the Personalization project and A/B test, which saw statistically significant (p<0.01) increases in CTR and read time for Scribd subscribers
• Worked on a faceted-search EBR system to power multiple recs services and scale for large (>10^8) corpora (Vespa)

• Elasticsearch
• Apache Spark
• AWS
• Scala
• Python

• Pioneered a neural network to estimate CPU utilization of Facebook live video transcoding requests using encoded parameters
• Implementing a dynamic slotting dispatcher to allocate optimal number of cores to requests, increasing server efficiency by 35%

• Python
• SQL
• C++
• Bash

• Built “More Like This” feature for The Roku Channel, which uses matrix factorization to personalize movie recommendations
• Employed Apache Spark and AWS Redis to build data pipelines, capable of handling user requests with <10ms latency

• Apache Spark
• AWS
• Scala
• Java

Netflix Workshop on Personalization, Recommendation, and Search (2021)

• Presented a poster discussing building an embedding-based retrieval system to serve recommendations with Elasticsearch
• Answered questions from and networked with machine learning practitioners across various industries

• Elasticsearch
• Apache Airflow
• Apache Spark

Finalist Project

Cross-Corpus Recommendations (Discovery Hack Week)

• Leveraged Gracenote metadata to prototype ML system capable of recommending D+ shows from a non-D+ query
• Achieved finalist status and presented demo to CTO and other executive members

• TensorFlow
• Docker
• AWS

Swipe to Discover

• Prototyped Tinder-inspired feature for users to swipe on books and generate preference signals (Scribd hack week)
• Built architecture to compute and retrieve recommendations in real-time using Elasticsearch and AWS Lambda

• Python
• AWS
• Elasticsearch

xterns

• Developed a platform designed to connect diverse, talented students from across the country to top employers through externships
• Built backend using Flask and MySQL, constructed internal analytics platform, and conducted A/B testing for the website

• Python
• Flask
• SQL
• HTML

Placed 2nd

Facebook Data Challenge

• Extracted and joined public Yelp review data with local business data to identify innovation opportunities in the restaurant market
• Presented proposal and data-driven approach to a panel of Facebook Data Scientists and Data Engineers

• Python
• SQL
• Bash

Microsoft Azure Finalist

DubHacks

• Built a webpage viewer to better the browsing experience for individuals that identify as Autistic or possessing developmental learning disabilities
• Employed Microsoft Azure's Cognitive Services API to automatically generate captions for scraped images

• Python
• Flask
• Microsoft Azure
• HTML

Image Colorization Algorithm

• Developed a convolutional neural network capable of colorizing grayscale images realistically
• Presented research to technical and nontechnical audiences at annual Northwestern research exposition

• Python
• Keras
• TensorFlow