I was responsible for revamping and redesigning the Showtimes feature for Siri where a user can ask Siri about theaters nearby. I was responsible for the new workflow and implementation of the feature along with developing a new UI also compatible with the Siri Jr. interface. My code went into the Siri pipeline at the end of my internship. I also presented my work to the head of Siri Experience who was impressed that I developed a shippable product at the end of 10 weeks of my 12 week internship.
Worked on projects in Machine Learning and Natural Language Processing. Created a probabilistic Gamma-Poisson classifier for record linkage which has now been taken into production. Also worked on supervised topic modeling and attribute-metric extraction to automate compliance checking of ESG reports.
Worked with Free UK Genealogy where I developed a framework that performs Named Entity Recognition on data present in wills and probate books from 18th century England.
Created Machine Learning and Deep Learning models for classification on a highly skewed dataset of identity cards
A collection of articles, presentations or published articles.
My team and I built a summarization system which extracts sentences from multiple documents that best represent the topic as its summary. This was a part of my LING 573 course on NLP Systems.
As a part of my LING 575 course on Neural Language Models, my team and I built a Named Entity Recognizer using BERT and ELMo
My final year undergraduate project which aims at automating question generation for images and verifying the result using Visual Question Answering techniques. Thus, jointly learning through these two tasks can bring improvements to Visual Question Generation (VQG) and also solves a significant issue of dataset annotation. The thesis proposes to integrate a VQG module with a VQA architecture to give an end-to-end architecture that performs self-evaluation of questions by creating QA pairs.
Built a Gamma-Poisson graphical model to present a supervised solution for fuzzy matching of record pairs aka record linkage.
See all Creations for more examples!
Built as a student developer for Google Summer of Code for the organization - Free UK Genealogy.I created an end-to-end system that could extract the text from probate books and seed them into a database with entities such as name, county, date, relationships, etc.