Jasper Zhou Email and Phone Number

Machine Learning Application Platform• LLMDeveloped an intelligent data Q&A system based on text2sql and RAG from scratch, achieving state-of-the-art results on public datasets, with over 95% accuracy in answering real databases containing more than100 tables;• Inventory Management OptimizationBuilt an inventory distribution simulation system from the ground up, simulating real demand and lead timedistributions, reducing the gap between simulation results and actual service levels to within 5%; Improvedmanufacturing scheduling efficiency, shortened retail replenishment cycles, increased replenishment algo-rithm frequency from 2.5 months to bi-weekly iterations, reducing inventory costs by approximately 7%;• Machine Learning Data PipelineEstablished an ML data pipeline from scratch, developed a proprietary data update management framework,supporting offline and online updates for over 100 ML tasks;• A/B TestingBuilt an A/B testing platform from the ground up, established a multi-layer experimental architecture, in-troduced techniques such as re-randomization and hash pre-warming, significantly improving experimentiteration efficiency and reducing experiment time by approximately 50%;

Machine Learning Platform• Feed Stream RecommendationResponsible for the development, optimization, and modular effect evaluation and testing of cold-start al-gorithms and diversity strategies;• Personalized Push NotificationsBuilt a business evaluation metrics system for personalized push scenarios from scratch, established anevent-based data flow framework, supporting services like TikTok live push notifications;• A/B TestingPromoted and developed uplift analysis techniques in causal inference, implemented in over 10 real-worldscenarios;

Industry Solutions• Banking Risk Control and Anti-Money Laundering SolutionsLed the design of data pipelines and feature stores for risk control and anti-money laundering scenarios,achieving an average recall rate improvement of over 40% compared to business benchmarks, deployedsolutions for more than 10 key account customers, and conducted over 50 customer training sessions;Machine Learning Platform• Automated Feature Engineering (AutoML)Developed automated generation and selection of multi-hop and temporal features for tabular data, sig-nificantly improving the development and deployment efficiency of financial ML applications, reducingdevelopment manpower time by over 50% under the same metrics;

Carpool Order Dispatch• Optimization of Carpool Order Dispatch AlgorithmsDeveloped model-based pre-matching strategies, reducing order dispatch service time by 15% within a0.1% decrease in order success rate; added long-term revenue models for orders, increasing the overallcarpool success rate by 1.2%;• Development of Order Dispatch Simulation SystemBased on real DiDi order data, developed an event-driven order dispatch simulation system, achievingestimations of key metrics such as carpool success rate and average income per person within 10% ofactual statistics, assisting in the optimization and iteration of online order algorithms;

Long Video Recommendation• Improved CVR through feature engineering in personalized recommendation and related recommendationscenarios for long videos; optimized creator incentives;