Multi-modal learning, vision-language models (videos), embodied AI

- Researched vision-language models for long-form videos.- Conducted large-scale, distributed training of vision-language models on video-text data using SageMaker.- Ran extensive evaluations on the current SOTA vision-language models.

- Developed a novel algorithm (MARDI) to train a unified named entity recognition (NER) model from multiple resources with different tag sets.- Architected an internal natural language processing (NLP) library that streamlined NLP model deployment to production.

- Refactored a monolithic server into microservices communicating over protobuf-based RPC calls and an AMQP message queue to scale real-time chat message delivery and support more resilient websocket connection handling.- Improved the deployment system to deliver true continuous integration and deployment.- Developed and deployed a set of microservices to support the “omni-channel” chat experience where customers can chat using their preferred communication channel, e.g. Apple Business Chat, push notifications, SMS, etc.- Conducted research in proprietary, multi-lingual customer intent classifier.- Mentored junior developers through the onboarding process.

- Software engineer on the Payments and Checkout team.- Developed and launched Guest Checkout, a critical project that added millions of dollars to Etsy's global merchandise sales.

- Developed and maintained an API written in Scala (Finch, Finagle) that allowed mobile apps to make purchases using Google In-app Billing.- Extended a Rails API to add new features such as in-app authentication.- Wrote and deployed a number of big data analytics job using Celery, RabbitMQ, Spark and HBase.- Developed and maintained an internal web app (Scala Finch, Postgres, Slick) that allowed employees to tag images with predefined characteristics to construct a data set that would be used to train a machine learning model.

- Implemented a Kubernetes-like system to integrate Docker to AppNexus’ cloud system.- Developed, automated and maintained a system that rapidly generates clones of the production system that can be used as sand boxes by developers and external clients.- Developed a proprietary configuration system similar to Puppet’s Hiera.- Managed a summer intern to successfully complete his project.- Developed and maintained tools to manage AppNexus’ cloud system, including a job scheduler and a server management web application. - Heavily involved in AppNexus’ recruitment process through interviews and on-campus recruiting events.

- Oversaw the technological needs of the eating club with a total of 150 members- Designed and implemented a web application that streamlined the club admissions process of my eating club- Implemented a user registration system with a smooth interaction between prospective members and current members, which is all backed by a PostgreSQL database- Used Python Django, Javascript, Twitter Bootstrap, Postgres SQL database, Heroku and Amazon S3

- Holds 1st degree Black Belt- Assist instruction of classes at the Princeton Judo Club located in a local YWCA

Supervise and teach classes for this student-run club with a total of 20 members

- Developed internal web tools, which increased information availability and efficiency of Facebook employees- Optimized internal search by improving search indexes- Migrated the old internal news feed code to the new code using Open Graph- Utilized the full stack of the Facebook code base: data access layer abstractions, PHP server-side code, HTML, CSS and Javascript

- Contributed to the authoring, testing, and operationalization of over 30 unique web crawling and parsing scripts in Python- Designed and implemented an enhancement to the core Postgres-SQL database schema to efficiently represent entities- Built intuitive, prompt-based programs for navigating through complex data, e.g. Google Analytics data