University of California, Berkeley

University of California, Berkeley, Haas School of Business

Machine learning research Scientist at Amazon - Alexa.

I teach Statistics and Machine Learning in the the Berkeley Master's in Data Science (MIDS) program.

Part time Lecturer

Leading a team of scientists and software engineers working on scalable generative AI techniques to increase customer interaction with their Alexa devices through improving their conversational experiences.  Specifically, I devise strategic roadmap for product release, work with various teams to scope requirements, do research and build the prototype technical solution, and oversee the end-to-end implementation in production.

Amazon

Senior Applied Science Manager, Amazon AGI (Artificial General Intelligence)

I was charged with developing and releasing new features for Alexa mobile devices, such as Echo Buds.  The work required partnering with various teams to ensure the end-to-end pipeline is setup and runs smoothly.  A recent new features includes: find the location of an object using a tracker device, and Fitness calories tracker.

● 	Conducted research and experimentations with applying federated learning framework to natural language understanding (NLU) models. The goal of the research was to deploy NLU models on edge devices to increase customer privacy. 
● 	Completed performance improvement for deep neural network (DNN) models used in the natural language understanding pipeline.  Methods included experimenting with different DNN architectures, and use of semi-supervised learning.
● 	Managed research and feature development for conversational experiences.

Senior Science Lead, Alexa Devices

As a Lead Machine Learning Scientist, I conducted research and developed the code and REST APIs (offered on AWS API Gateway) for: topic modeling, text summarization, document recommendation, sentiment analysis, and streaming text analytics.

● 	Engaged with prospect clients to develop proof of concept for their needs.
● 	Prepared various datasets for use with the software, such as twitter feeds, financial market reports and research, and legal documents.
● 	Developed unit and integration tests for the APIs to test for accuracy and speed.
● 	Partnered with frontend engineers to implement the full software pipeline.

SumUp Analytics

Lead Machine learning Scientist

●	Led a variety of engagements for banks and financial institutions, as both quantitative model lead and team manager.  The projects covered a spectrum of quantitative risk measurement products and services.  Example projects included: Credit risk modeling for commercial wholesale portfolio, Commercial Real Estate, Residential Mortgages, and Counterparty Credit Risk.

Moody's Analytics

Director of Data Science and Machine Learning 

I participated in supervisory exams as the lead quant analyst on bank holding companies regulated by the Federal Reserve Bank of San Francisco including Wells Fargo, Union Bank of California, and Zions Bank. I also interacted extensively with other examiners and bank modelers and reviewed various credit loss forecasting models, stress testing models, and economic capital models regarding the models’ conceptual soundness, implementation quality, performance evaluation, and validation practice.

● 	Participated in system-wide projects within the Federal Reserve such as the Comprehensive Capital Analysis and Review (CCAR) and BASEL II quantification exams.
● 	Executed empirical research on credit risk modeling, systematic risk indicator using Credit Default Swap (CDS) spreads, and Commercial Real Estate (CRE) modeling and analysis. The research included: data gathering using Bloomberg, RCA, FRED, FR Y-9C and 14A bank schedules, data cleaning, building time series models, analyzing the results of model forecasts
● 	Completed research on alternative measures of risk for portfolio optimization.

Federal Reserve Bank of San Francisco

Senior Quantitative Risk Analyst

● 	Managed the development of statistical risk model for the emerging markets using principal component analysis (PCA).
● 	Engineered a strategy based on the minimum variance portfolio and testing the strategy which led to the creation of AllianzGI U.S. Managed Volatility Fund (NGWAX).
● 	Created an expert system for risk management, change point analysis, regime-switching GARCH models, Goodness-of-fit tests for copula, and application of Kalman filter to Fama French factor model.

Tanya Roosta