Wearables AI Infrastructure

Privacy Infrastructure

• Only individual in 82-person org to receive highest-possible performance rating of “Redefined Expectations” in 2023; again in 2024.

Siri: Natural Language Understanding

• Developed NLU and generative models in TensorFlow to improve multilingual Siri parser accuracy for on-client domains, with a focus on scalability to low-resource languages. Designed, implemented, and deployed NLP software package for few-shot synthetic data generation of PII-free conversational language data, used by Siri teams for multilingual model training.

• Implemented language models end-to-end with data pipelines, training pipelines, on-device testing. Used Docker, AWS S3, protobuf.

Natural Language Processing Lab: Language, Information and Learning (LILY) under Dr. Dragomir Radev

• Collaborative project with Facebook AI. Researched and developed NLP models for hybrid extractive–abstractive text summarization and argument graph modeling.

• Created conversation summarization benchmark on 40,000+ dialogues by implementing clustering and summarization pipeline with PyTorch, HuggingFace Transformers, sklearn. Used pandas, NumPy for data processing. Ran on Linux, used git.

• Improved support agent efficiency in 14+ countries by developing tools for admin-level user authorizations and product pricing.

• Developed year-in-review statistics feature for 214,000 paid users, including database backfill and user-facing modals.

• Coded frontend with React, Redux, JavaScript/JSX, jQuery; backend and database with Python, MongoDB, pymongo. Used Git, Jira, Unix, CircleCI.