Mountain View, California, United States
• Gemini Omni post-training core contributor. • Veo 3.1 post-training core contributor. • Imagen 4 post-training contributor. Core contributor on image reference capabilities. Gemini Nano Banana contributor on data workstream. • Talk on video generation model works at YouTube at the Google Booth at ICCV 2025. • Veo 1 core contributor. • First-author paper at ECCV2024, on multimodal visual understanding. • Core contributor to internal large video-text dataset.
• Trained Light GBM model for driver arrival time prediction. Reduced prediction error by 14% compared to current baseline. • Implemented online model inference pipeline in Golang. • Launched an A/B test in production to test the effect of displaying to users improved driver arrival time estimation. • Launched an A/B test for experiment selectively nudging users away from requesting during driver undersupply if pickup wait is too long compared to actual ride time, to increase overall efficiency by minimizing unproductive time getting to riders.
• Implemented and ran various baseline experiments in PyTorch. • Trained unsupervised visual scene segmentation model and assisted in implementation for paper accepted at ICCV 2021.
• Built database-integrated Python framework for automated metrics projection using statistics models. • Designed algorithm to solve pagination drifting issues in API calls. • Created demos for Teradata analytics platform’s new ML engine.
• Trained deep learning model to classify neuron signals. • Built automated GUI and ETL pipeline for data-labelling.