Jeff Sontag Email & Phone Number

United States

Durango, Colorado, United States

- Pioneering the application of latent embedding techniques for audio, enabling open source music generation and categorization for artists and developers- Proposed, trained, and released novel Jukebox Diffusion framework for generating high-quality, 75-second music clip generations from noised latents- Published paper in Better Programming to 45k shares, 3k reads, 100 Github stars

Santa Monica, California, United States

- Led product research team of 10 Data Scientists, Product Managers, and User Researchers analyzing thousands of user behavioral features to identify product gaps- Proposed original unsupervised clustering technique to segment users by behaviors and evangelized clusters across team for acquisition targeting and new product dev- Developed multi-head regression model for transparent customer facing smart recommendation product, tested new feature at record 68% customer engagement- Launched smart recommendation model live in Kenya market, increasing early product customer retention by ~5%

Santa Monica, California, United States

- Built predictive models end-to-end for real time credit scoring and offering of 5 million customers more than $2 billion in loans- Improved feature engineering techniques for DS team by training deep learning models to embed thousands of sparse Android data points in frozen training vectors- Introduced innovative validation set practice for DS team, optimizing for through time model stability evaluation, improving average lifespan of models by ~3 months- Independently trained and deployed Tala's inaugural predictive lending model in Mexico, facilitating over $12M in lending through approximately 800k decisions

Santa Monica, California, United States

- Managed event driven behavioral data collection with Amplitude, defined event triggers with Product Managers and ensured Engineering implementation- Consolidated data access into standardized Amplitude and Looker dashboards, maintained organized documentation for event analytics across all of TALA- Investigated application of behavioral data signals for use in fraud detection, identified 8 features with significant predictive power, decreasing loan fraud by 35%- Decreased applicant churn by 6% through statistical design of A/B tests to address product drop-off points

Denver, Colorado, United States

Denver, Colorado, United States

Los Angeles, California, United States

Master'S Degree, Business Analytics, 4.0 Gpa

Bachelor Of Science - Bs, Business Administration And Management, General

Activities and Societies: Trustee Scholarship - Full Tuition Academic Scholarship