robinwu8080

Tech Stack: Coral (Amazon’s Java Framework), FastAPI, LangGraph, DynamoDB, OpenSearch, Ripple, Cradle,React Project: VAMOS Scorecard — Real-time vendor performance dashboard for Amazon Japan, integrating 60+ core retail metrics (Revenue, Net PPM, etc.) with multi-dimensional analysis and AI-driven insights, serving all vendors • Designed and implemented a tool-driven Vendor Performance Insight Agent using LangGraph. Based on metric performance, time range, and intermediate findings, the agent dynamically selected Trend, Goal, Benchmark, and Contribution analysis tools and drilled down into Category, Subcategory, Brand, or ASIN dimensions as needed, generating structured Insight Cards with goal gaps, top and bottom contributors, and actionable recommendations • Implemented deterministic calculations for metric values, goal gaps, trends, and contribution ratios, and validated all numerical and contributor claims against tool outputs. Introduced bounded replanning, duplicate tool-call detection, execution timeouts, and rule-based fallback templates to prevent infinite loops and unsupported conclusions • Built end-to-end agent observability to trace LangGraph node execution, LLM requests, and tool calls, capturing latency, token usage, and fact-validation results. Integrated Prometheus and Grafana to monitor insight-generation success rate, end-to-end P95 latency, and model and tool error rates • Designed and implemented a Cradle-based ETL pipeline to extract pre-aggregated metric data from Ripple and batch-load it into OpenSear

Tech Stack: Spring Cloud, MySQL, Redis, RocketMQ, Kubernetes, Prometheus Project: Xibo Education Flash-Sale System — Weekly time-limited book-grab campaign with peak QPS of 20K, driving significant user acquisition for the education platform • Led architecture design and development of the core flash-sale module; implemented atomic Redis inventory deduction via Lua script to prevent overselling; leveraged RocketMQ for traffic peak-shaving, supporting 20K peak QPS during the events • Designed risk-control module: maintained UserID blacklist with Redis Set; implemented request-rate limiting via 10s sliding window with atomic INCR counters; investigated and blocked bot-driven order fraud • Built multi-layer monitoring: tracked Pod CPU throttle rate via Kubernetes cAdvisor; monitored JVM thread states with Actuator + Micrometer to detect lock contention and I/O bottlenecks; monitored Redis command P99 latency and memory usage to prevent key eviction from corrupting inventory data • Designed Redis failure fallback: configured connection timeout threshold; auto-disabled the entire event entry after consecutive inventory Redis timeouts with alerting; gracefully degraded risk-control Redis failures to pass-through without affecting sales; periodic health-check probe with manual confirmation before re-enabling the Redis instance to avoid flapping

Tech Stack: Kubernetes, Spring Cloud, Consul, MySQL, Sharding JDBC, StarRocks, Kafka, Prometheus Project: Sam’s Club Omnichannel Replenishment System — Multi-scenario demand forecasting model that auto-generates SKU-level replenishment orders and plans • Led performance optimization of the replenishment calculation pipeline by decomposing large batch jobs into parallel tasks and tuning worker concurrency, reducing batch computation time by 60% • Designed and implemented a Sharding-JDBC table-sharding strategy to handle 15M+ new records per day, improving read/write performance for high-volume replenishment plan data.Built the supply chain team’s Prometheus monitoring system: designed Grafana dashboards and alerting rules, packaged Micrometer monitoring utilities for cross-team reuse, and delivered team-wide monitoring tech talks • Diagnosed and resolved multiple critical production issues (Page Cache inactive file inflation from disk-based logging causing cAdvisor-reported container memory to exceed limits and trigger frequent restarts, misconfigured health checks triggering sporadic pod restarts, and database deadlocks) producing systematic troubleshooting documentation adopted across teams

Tech Stack: Spring Cloud, Kafka, ElasticSearch, MySQL, Sharding JDBC, Redis, Kubernetes Project: Intelligent Delay Management System — Processing 5–10M daily delay tickets across 50–60M daily shipments, with rule-engine-based automated accountability assignment • Developed core modules and implemented a Sharding JDBC-based sharding strategy for 5–10M daily delay work orders • Introduced ElasticSearch as a query optimization layer over Sharding JDBC-partitioned MySQL data, significantly improving deep pagination and bulk export performance; implemented RocketMQ delayed queues for automated work-order timeout state transitions, diagnosing and resolving message backlog issues

Tech Stack: Spring Cloud, MyBatis, MySQL, Redis, Kafka, React Project: ShipperBee Delivery Platform — Crowdsourced distributed logistics platform supporting P2P and relay delivery modes, solving last-mile delivery efficiency bottlenecks • Canary Release System: Led the design and implementation of a full-link canary release system based on traffic coloring, enabling precise traffic routing from gateway to downstream services and supporting user-level feature validation with zero-downtime deployments • Microservice Governance: Fine-tuned Spring Cloud Eureka registry parameters (heartbeat detection, self-preservation), resolving service state synchronization delays caused by frequent registration/deregistration, reducing service discovery latency by 50%+ • Core Business Development: Owned the pricing engine and exception handling center; optimized MySQL slow queries and index structures, reducing core API response time by 30% under high concurrency

Tech Stack: Spring MVC, Spring, MyBatis, MySQL, Tomcat • Responsible for business module design and development, including database schema design, API development, and unit testing