Gary Song Email & Phone Number

New York, Ny, Us

New York, United States

New York, United States

Part of a three-person pod with a Sr. Staff Machine Learning Engineer, did EDA using PySpark on an EMR cluster to understand user behaviour, inform feature engineering and selection; Worked on Airflow data and model training pipelines that run daily; Worked on serving model via microservice and monitoring via DataDog dashboards; Troubleshooting with SplunkInvestigated ingestion pipelines and codebase to recover 96% of data necessary for much more accurate impressions count, a key metric that was very inaccurate up to that pointAs an ask by the SVP of Software, worked with Head of Product to redefine metrics for monitoring and improving the health and efficiency of codebase, deployment workflows, and identify technical debtAnalyzed data to support effectiveness of onboarding program and enforce best practicesPart of working group for expanding developer insights to include infrastructure insightsParticipated in design and adoption discussions regarding tech stack (feature store, modelling framework, etc)Produced design docs and LucidChart diagrams for changes to data pipelines, data flow, etc

Scoped and started building new content recommender for the Unity Learn website to drive user engagement with greater goal of establishing a capability for recommender systems as part of the team’s roadmapTechnical lead for churn prediction project for the core business, oversaw delivery of two model iterations managing multiple stakeholders across three time zones – All parts of the data science life cycle, including EDA, feature engineering, model building, model selection and validation, and product deployment as end-to-end pipeline with KubeFlow; 25% increase in PR curve AUC vs. previous modelExpanded churn prediction project scope to cover managed accounts, thus accounting for all editor accounts across the companyBuilt first iteration of self-supervised deep learning models on behavioural data for the user segmentation project, which was used to inform the Board of Directors of the changes in user-type composition of Unity engine user cohortsRan A/B test to assess impact of existing asset store recommendation modelWorked with the data infrastructure team to establish best practices for end-to-end machine learning pipelinesOnboarded and mentored junior team members

Mississauga, Ontario, Canada

Designed, implemented, and maintained web scraping engine for automated data collection at scale of 44+ million data points monthly, used for competitor pricing analysis for bid support; In addition to saving $22,000+ monthly, the data collected allowed our team to support $1+ billion in contract bids across all of USA in 2019 alone, resulting in $30+ million liftDeveloped system that uses deep learning and classical computer vision to automate inventory control in near real-time; Due to be piloted in partnership with $2+ billion subsidiary of Compass Group USA, CDL’s parent company; Demoed at CX trade show for the Canadian business and site operatorsOrganized and led bi-weekly workshops for Data Science team, with goal of ramping up team’s skills to state of the art in deep learning, computer vision, NLP, web scraping, etcDesigned, implemented, and maintained pipeline and model for semi-supervised tagging of internal client data; Cleaned data was used downstream in almost all analytics, including CDL’s Advanced Analytics team’s case studies and in CDL’s flagship data productSuccessfully advocated for department to branch into engineering, on basis of increasing data ownership, product ownership, speed to market, etcSuccessfully advocated for formalization of R&D initiatives, leading to establishing of machine learning lab

Mississauga, Ontario, Canada

Proposed methodology for inferring customer characteristics based on cellphone GPS location dataSupported case studies via web scraping, doing geospatial analysis, and data mungingPerformed data analysis for various subsidaries of Compass Group USA, CDL’s parent companyAs one of the first hires, worked closely with the team to brainstorm for new initiatives to raise the group’s visibility and establish the group’s reputation, in order to grow the team

Bachelor'S Degree, Statistics

Coursework, Mathematics