Jun 2021 – Jun 2023 · 2 yrs · Menlo Park, California, United States

• As part of the Integrity team, developed ML classifiers to scale health misinfo detection across Facebook & Instagram, resulting in 17% Prevalence reduction or ~96 Million misleading posts filtered/day.
• Experimented with several model architectures (XGBoost, Neural Networks) by performing transfer learning on Textray & other proprietary embeddings.
• Led model consolidation efforts across 19 languages saving compute by 50%.
• Built shared job routing infra for Content Moderation, saving over $2 M.
• Built dashboards for the 2022 Brazil and US midterm elections. Used BERT Topic Clustering to surface civic and voting misinformation.

Sep 2020 – Dec 2020 · 3 mos

• CIS-519 Applied Machine Learning

May 2020 – Aug 2020 · 3 mos

• Worked as part of the Ads Core ML Modeling team to develop a cost aware method for feature selection. This project helped to improve the efficiency of ads ranking models via reduced CPU, Network and Memory usage.

Jan 2020 – May 2021 · 1 yr 4 mos · Philadelphia, Pennsylvania, United States

• Worked towards developing SHAP explanation plots for PennAI, an open source analytics tool built by the Institute of Bioinformatics. SHapley Additive exPlanations are a recent approach towards the interpretability of machine learning models. My contributions involved applying Kernel and Tree explainers for sklearn-based models in the backend, as well as developing react/redux ui for showing the generated summary plots in frontend.
• More about PennAI - It is a docker microservices project that contains a custom python autoML recommender engine, a node backend for performing RESTful communication between micro-services, and a react/redux ui.

May 2019 – Aug 2019 · 3 mos · New Delhi Area, India

• As part of the Web Interoperability Team, I was responsible for writing a Rust Library for the Marionette Remote Protocol. This Rust library was used as an API by the Geckodriver to interface the HTML Request and Response messages with the Firefox server.

Jul 2018 – Sep 2018 · 2 mos

• Worked with Mozilla as a part of Rails Girls Summer of Code programme!
• (An open source development initiative)
• My work involved coding in Rust on Servo, the browser layout engine.
• Added support for Responsive Images by implementing several algorithms such as update source set, react to environment changes, select image source, etc.

Aug 2018 – Dec 2018 · 4 mos

• Worked in the Multimedia Domain, scraping text, audio and video content from Twitter in regards to the 'Kiki Challenge'. This dangerous dance challenge had claimed several lives due to people jumping out of running cars and dancing on the road.
• Developed a Deep learning pipeline to classify the video as safe/unsafe.
• Used transfer learning to experiment on models such VGG, ResNet.
• Data Augmentation was also considered to ensure dataset is balanced.

May 2018 – Jul 2018 · 2 mos · Bengaluru Area, India

• Developed methods for formal verification of software in Java.
• Got to learn a lot about Coq and functional programming
• Creating Abstract Syntax Trees
• Equivalence of Programs