* Engineered a real-time fraud detection platform powered by AI & ML models, integrating low-latency APIs for real-time detection and Flink & Spark pipelines for high-volume batch processing

* Achieved 15,000 QPS at < 100ms latency across a multi-cluster distributed, high-availability architecture by using Kafka to decouple microservices and synchronize data for real-time consistency

* Achieved 95%+ accuracy and 30% reduction in go-live time by orchestrating prompt engineering (Chain-of-Thought, Tree-of-Thoughts etc.) for a LangChain agent to translate language into fraud-detection entities

* Boosted env migration efficiency 5x and reliability to 99.6% from 90% with flexibility by re-architecting an import/export system, implementing multi-threaded concurrent processing to handle complex dependencies

* Delivered 99.99% message reliability for 1M+ Kafka events/d by refactoring the consumption model with deferred commits & circuit breaker, along with tiered retry MQ & S3 DLQ for downstream services

* Improved computing efficiency and 20% resource utilization by refactor feature aggregation engine, optimized PostgreSQL query by multi-level pre-aggregation cache buckets for hotspot and offline computing for batch