Haitao Liu

Senior software engineer with extensive experience in architecting and developing online streaming and offline batch services on distributed systems utilized by machine learning applications

China National Scholarship, 2010

• Google Play Store Incentive Offer Retrieval Platform, Redesigned the Incentive Offer Retrieval Platform by introducing a mini runtime workflow framework, which allows config driven plugins to enable various capabilities including User Targeting, Content Personalization, State Persistence, Governance Snoozing, Abuse Prevention, and Rule based and ML based Ranking. Successfully onboarded multiple serving flows and landed important use cases on Store Surfaces requested by different Vertical teams (Games, Apps, Search, Loyalty, Commerce teams, etc.)
• Self-Service Modeling and Bulk Scoring Platform, Led a team and developed a Data Analytics Machine Learning Platform and Scoring Pipeline, supporting Camel Connector Importer, data Matching and Enrichment, Cascading data Transformation, Analytics/Scoring pipeline and data Export. Designed a Kafka based distributed Job Scheduling & Orchestration platform. Implemented the Service Registry Watcher, Distributed Lock and Caching layer using Redis.
• Hadoop Yarn Data Platform, Architected the Container Allocation framework on top of Hadoop Yarn to allow running different types of jobs with different resources. The resource efficiency had been improved by 30%., Added the support for Resource Manager HA in spring-yarn-core library, by performing failover on top of regular RPC Proxy.
• Machine Learning Modeling Pipeline, Worked on the Supervised Machine Learning Pipeline for feature generation in Python, involving Null Value Imputation, Discrete Value Bucketing, Embeddings, etc, and building Random Forest Models. Output Model evaluation Scores and Feature Importance for Visualization on UI.
• Prediction Engine, Redesigned the batch-processing prediction engine with a more light weight, multi-tenant, supporting both Quartz scheduled and on demand requests based and distributed version, by leveraging the MapReduce framework with 15x throughput improvement.

• Neural Networks and Deep Learning
• Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization
• Sequence Models
• Generative AI with Large Language Models