• Designed and deployed a transformer-based demand forecasting system on AWS, integrating historical sales, seasonality signals, and external variables; improved forecast accuracy by 15% and reduced stockout incidents by 20% across supply chain operations

• Built a production-grade retrieval system for regulatory and compliance documents using hybrid search (dense + lexical), reducing document lookup time by 40% and significantly improving decision turnaround for compliance teams

• Engineered high-throughput model serving infrastructure using Kubernetes (EKS) and optimized inference via batching and GPU acceleration, increasing real-time inference throughput by 10x+ under peak loads

• Developed end-to-end MLOps pipelines with automated training, validation, and deployment using MLflow and CI/CD workflows, reducing model release cycles from weeks to days and improving deployment reliability

• Implemented distributed data pipelines using Spark and Kafka to process ~8M daily records, enabling near real-time feature generation and reducing data latency by 50%s

• Built and deployed an LLM-powered customer support assistant using retrieval-augmented generation, improving query resolution accuracy and increasing customer satisfaction (NPS) by 12%

• Developed a fraud detection system combining graph-based embeddings and gradient boosting models, improving precision at high recall thresholds and reducing false positives in real-time scoring pipelines

• Designed online experimentation frameworks, including shadow deployments, canary releases, and threshold calibration to control false positive rates below 1%, reducing production risk and rollback frequency

• Standardized feature engineering pipelines and introduced reusable feature storage on distributed systems, reducing model retraining cycles and improving consistency across fraud and risk models

• Developed churn and credit risk models using gradient boosting techniques with engineered temporal and behavioral features, contributing to measurable reductions in churn and high-risk approvals

• Built recommendation models for next-best-offer systems using candidate generation and ranking approaches, improving conversion rates in A/B testing experiments

• Optimized ETL pipelines with incremental processing and validation checks, reducing runtime and improving reliability of downstream analytics systems