Shopping Ads Infra Team

Shopping Ads Serving Infra team

Search Ads Quality Serving team.• Worked on ads serving infra to ensure production safety and improve developer velocity.• Modernized search ads server through applying a next-generation programming model.

Ads Deep Linking in Google Ads• Worked on a full-stack project to delivered frictionless mobile ad experiences across web and apps.• Developed an internal tool across distributed systems to simplify authenticated access to log data.• Built web app and dashboards for sales team to understand and act on the status of customer setup.• Worked on training model to predict app uninstalls based on user activity.

• Worked as tutor and grader for Algorithm, Computability and Logic, etc.• Worked as grader for Intermediate Probability, Calculus, Special Relativity, etc.

• Completed a deep learning bootcamp (fast.ai) to learn topics in machine learning (ML) and neural networks to implement a ML solution to the live song and score identification problem.• Worked on data scraping and metadata construction to build a system for sheet music identification.• Performed research on state-of-the-art approaches in instance retrieval using deep learning.

OCR in Google Research and Machine Intelligence• Developed optical character recognition (OCR) models for historical manuscripts, improving character error rates by 10-60%.• Developed tools in C++ and Python to process external public datasets (from >5 sources) into format used to train deep neural network OCR models in Tensorflow.• Improved OCR model by 30% on historical Arabic dataset by training language models, developing rendering tools to simulate ink bleed-through, and conducting data augmentation experiments.

• Investigated image recognition of text that is eroded, dilated, or unreadable.• Reduced the analysis time to 1 second per text image with almost 100% accuracy.• Quantify the level of erosion and dilation using Gaussian distribution level.• Presented findings to company in weekly conference calls and wrote final report.