Core ML Engineering Productivity

Head TA for CS 170 Algorithms and CS Theory Aug 2021 - Present; Head TA for CS 70 Discrete Math & Probability Theory June 2021 - Aug 2021 In charge of course-running, course staff management and administrative matters of a 700-student class. Regular meetings with course staff and professors. Spearheaded multiple facets of course-running; including exam-writing, content creation, coordinating logistics, syncing with the professors, website feature creation, etc.

Advised by Prof. Dawn Song. Research on Privacy-Preserving Data Science. CoLink provides a unified interface for the user, storage, communication, and computation. Extending gRPC, CoLink simplifies the development of multi-party protocols and allow implementations in different programming languages to work together consistently. Skills: Rust (Programming Language)

Intern for the internal A/B testing framework team at Google. Created a library that computes the validity of queries via a new SAT solver - Cooperating Validity Checker (CVC) - using Go and C++. Provided detailed program verification analysis and explanation of invalid queries in 384385 A/B experimentations by 43074 Google developers that use the experimentation framework via the library I built. Extremely scalable - no-preprocessing multi-threading design makes efficiency scale with the number of cores. Sped up diff analysis by 10000x through caching mechanism I designed and implemented. Skills: C++ · Protocol Buffers · Go (Programming Language) · Bazel · A/B Testing

Meta AI post about this project: https://ai.facebook.com/blog/assessing-fairness-of-our-products-while-protecting-peoples-privacy/ Designed and implemented a security protocol for machine learning evaluation fairness using secure MPC, homomorphic encryption, and zero-knowledge proof. Wrote 6000+ lines of Rust code that passes industry-level auditing. Built a scalable system that supports efficient fairness evaluation for a million users within one hour. First ever to achieve this using Secure Multiparty Computation. System deployed by Meta. Biggest deployment of Secure Multiparty Computation so far. Implemented the entire API and web server for the system, including code that performs cryptographic protocols, error handling, database CRUD actions, etc. Skills: Rust (Programming Language) · Firebase · Postman API · Google Cloud Platform (GCP) · Cloud Firestore

Summer internship at InitialView, a famous company for interview services, especially for international applicants for college or grad school. Implemented numerous new features on the Web App using Vue for the frontend and Django for the backend, including: a high school search feature for admission officers; virtual scrolling for search results; a student data input feature for counselors, and the department profile page for admission officers to view the student data, etc. Skills: Vue.js · Django REST Framework

Cross-Lingual Multi-Label Sentiment Classification with Adversarial Networks: Designed, developed, and evaluated a brand new, fast and efficient way for Sentiment Classification: a feed-forward network through a language-agnostic sentence representation model, and then through DNNs controlled by a language-adversarial network. Increased performance by 12% from the previously used Sentiment Classification model with the new approach. State-of-the-art accuracy in multiple languages. Skills: Machine Learning · PyTorch · Artificial Intelligence (AI) · Matplotlib · NumPy CrossLingual_Paper.pdf