Machine Learning Engineer
Current• Cost Prediction Pipelines: Designed automated pipelines for workers’ compensation and auto liability cost prediction using LGBM, KMeans, TensorFlow, and Optuna. Enhanced automation and interpretability.• MLOps Implementation: Built an end-to-end MLOps pipeline using AWS SageMaker, Docker, MLflow, and FastAPI, reducing deployment time by 30%.• Fraud Detection: Developed a comprehensive fraud detection pipeline combining supervised/unsupervised learning, graph-based anomaly detection, and text mining. Achieved a 2x increase in high-risk entity detection.• Risk and Event Prediction: Created time-series neural network models (RNN, LSTM) to monitor claims, detect high-cost trends, and predict high-risk events. Enabled early detection of 40% more high-risk cases.• Text Mining Frameworks: Designed scalable text mining pipelines for unstructured claim data, integrating LLMs (e.g., BERT, SBERT) and traditional NLP. Improved delivery speed by 50% and reduced costs by 70%.• De-Identification Protocol: Led the development of PII de-identification pipelines using SpaCy NER, AWS Comprehend, and Spark NLP, ensuring compliance with legal standards.• Model Optimization and Innovation: Applied advanced techniques like ensemble modeling, clustering, Shapley values, and unsupervised embedding to enhance prediction accuracy and system reliability.Performance Evaluation: Standardized validation systems for scoring medical and legal claims, improving consistency and transparency.Technical Expertise:• Machine Learning & NLP: TensorFlow, PyTorch, LSTM, RNN, BERT, SBERT, clustering, LDA, Optuna, LLM refining.• MLOps & Cloud: AWS SageMaker, Docker, MLflow, FastAPI, and CloudWatch.• Data Science Tools: Python, PySpark, ETL, and Shapley analysis.• Anomaly Detection & Text Analysis: Graph-based methods, topic modeling, and LLM-driven analysis.
