Liang Xin

Led development of scalable advertising services within Walmart Connect, building high-throughput backend systems and advertiser-facing tools to process 10M+ ad events daily, integrated with ML-driven optimization, maintaining 99.9% availability during peak retail traffic. Designed idempotent, high-throughput ad event ingestion services processing 50M+ daily impressions and click events, implementing Kafka-based asynchronous stream processing to decouple ad services from analytics workloads. Delivered agentic AI orchestration service that continuously analyzed performance signals (CTR, CVR, pacing, spend velocity) to recommend bid and targeting adjustments, reducing manual campaign tuning effort by 30%.

Maintained and enhanced the core authenticated domain supporting 10M+ active financial users, ensuring 99.9% availability across desktop and mobile browsers. Designed and implemented REST services in Node.js/Python to orchestrate LLM-driven workflows, integrating credit bureau data and behavioral signals with 2s response latency. Created backend orchestration services integrating rule-based financial logic with LLM-driven analysis, delivering personalized actions and reducing recommendation processing time by 38% through async inference and caching. Implemented RAG workflows integrated with real-time, event-driven inference pipelines to deliver contextual financial education and personalized AI recommendations, increasing user engagement by 25% and recommendation accuracy.

Architected an event-driven microservices architecture within the Classification Policy Platform in AWS Cloud, enabling asynchronous processing of billions of ASIN lifecycle events and improving workflow efficiency by 20%. Led the design and implementation of User Preference microservices in AWS, powering configuration management of 10K+ internal users for enterprise-scale compliance workflows. Delivered IDX platform combining React/TypeScript FE with event-driven Node.js/Java microservices on AWS, enabling lifecycle management of 10M+ catalog items through CI/CD release workflows with high scalability and reliability. Improved ML model training and deployment workflows for ASIN compliance classification, enabling scalable evaluation of 1M+ classification events per day to reduce model experimentation turnaround time by 30%.