Designed and deployed production-grade ML models for fraud detection and recommendation systems using Python, TensorFlow, and XGBoost,
improving model accuracy by 40% and reducing prediction latency by 50%.
• Architected real-time streaming pipelines using Apache Kafka, AWS Glue, and PySpark to process 2M+ daily transactions for near real-time risk
scoring and feature generation.
• Implemented scalable feature engineering pipelines, including feature extraction, hashing, normalization, and time-windowing, ensuring
consistency across batch and streaming systems.
• Built and automated end-to-end MLOps pipelines using AWS SageMaker, MLflow, and CI/CD (GitLab), reducing deployment time by 45% and
enabling continuous training and delivery.
• Containerized model serving using Docker and deployed RESTful APIs (FastAPI/Flask) on Kubernetes and AWS ECS with autoscaling and lowlatency inference.
• Developed and integrated model monitoring, logging, and data drift detection using SageMaker Model Monitor, Evidently AI, and Prometheus,
improving issue detection by 30%.
• Reducedfraud-related false positives by 25% through ensemble modeling, threshold optimization, and continuous A/B testing.
• Applied transformer-based NLP models (BERT) and embedding techniques (Word2Vec, sentence-transformers) to financial text, improving
recommendation relevance by 35% and user engagement by 45%.
• Developed GenAI-powered solutions leveraging LLMs, prompt engineering, and retrieval-augmented generation (RAG) for intelligent document
processing and contextual recommendations.
• Conducted rigorous model evaluation using cross-validation, ablation studies, and metrics such as ROC-AUC, precision, recall, F1-score, and
RMSE to ensure robustness.
•
