New York, New York, United States
Speech Research - New York (May 2020 - April 2022) • Built end-to-end automated speech recognition systems for Cloud customers. • Pushed ~10 ASR models in production, serving Cloud customers in various languages. • Worked with Google Brain to create state-of-the-art deep learning models for speech recognition. • Published 3 papers to leading speech conferences (https://arxiv.org/abs/2010.12096, https://arxiv.org/abs/2104.14346, https://arxiv.org/pdf/2110.03841.pdf) and filed 3 patents. Google Brain - Mountain View (July 2018 - May 2020) • Built NLP models that predict patients conditions based on medical reports. • Used NLP generated labels to predict infarct and hemorrhage from 3D images of the brain. Core Search - Mountain View (July 2018 - May 2020) • Built ranking models for Google Images. • Designed ML models used for triggering product features.
Machine Learning • Built a ML model that assessed the quality of Driver Profile Photos in real time. The model is live globally and has reduced errors due to bad photos from 42% to 1% -- Patent: https://patents.google.com/patent/US20190295240A1/en. • Built a prediction algorithm that leveraged riders' ticket data, drivers' GPS traces and smartphones' accelerometer data patterns, to predict fraudulent behavior on the platform. This included working on large and sparse data sets using Hadoop, Spark, R and Python. Product • Owned Uber's facial recognition product, which aims to ensure that the driver using the app matches the account on file. This included identifying ML opportunities, building data pipelines and testing the model through experimentation. • Worked cross-functionally with Product Managers and Engineers, tackling everything from estimating the feasibility of product launches in new markets to measuring their impact through A/B testing or by using causal inference methods such as regression frameworks and weighted-based models. Leadership • Led skill and knowledge sharing initiatives for Data Scientists in the company. • Created Spark and Deep Learning tutorials and organized tech talks. • Managed a Document Analyst.
• Taught UC Berkeley's Probability Course (STAT 134). This is a 300 person advanced undergraduate graduate course. • Held lab sessions, reviewed and introduced topics in probability, as well as office hours. • Designed and graded exams.
• Created a software that analyzed a patients walking pattern to predict their potential illness such as knee osteoarthritis. • Performed an automated heel strike and toe-off detection method for doctor’s office inertial sensor gait measurement (patent pending) • Used automatic detection of gait phases for feature extraction • Worked in partnership with the Applied Mathematic Center of ENS, the Cognition and Action Group Cognac-G, CNRS and Université Paris Descartes.