• Developed an AI-driven Fraud Detection platform analyzing millions of daily transactions, detecting anomalies and suspicious patterns, reducing potential fraud exposure by 35%, and improving risk response times while maintaining strict SOX and SOC 2 compliance.

• Built end-to-end data ingestion pipelines using Python, SQLAlchemy, and Pydantic, achieving 99.5% schema validation accuracy, with structured data stored in PostgreSQL, ensuring traceability, audit readiness, and regulatory compliance across large-scale financial transaction datasets.

• Engineered transaction-level NLP features including risk triggers, anomaly descriptors, and temporal sequences using spaCy and custom rule-based matchers, increasing input quality for downstream ML models by 25% and improving predictive classification performance.

• Fine-tuned Longformer and FinBERT models on 200K+ annotated transactions, achieving F1 score improvement from 0.72 to 0.88, leveraging domain adaptation, LoRA, and PEFT to reduce compute requirements by 40% without compromising security compliance.

• Designed real-time RAG pipelines using LangChain, FAISS, and AWS Lambda to provide accurate historical transaction retrieval, increasing query accuracy by 30%, while Streamlit dashboards reduced compliance analysts’ investigation time by 35% across workflows.

• Integrated scalable monitoring, evaluation, and deployment frameworks for ML models, ensuring constant performance tracking, automatic retraining, and full auditability, enabling faster internal deployment while maintaining regulatory standards.

• Assisted in developing AI pipelines to detect anomalies in healthcare transactions, using Python and NLP with FinBERT to extract risk-related features, improving data quality and supporting predictive classification models effectively.

• Built and validated data ingestion pipelines with SQLAlchemy and Pydantic, storing structured healthcare datasets in PostgreSQL, ensuring accurate schema validation, traceability, and retrieval for compliance workflows, while assisting with historical transaction analysis using LangChain.

• Supported fine-tuning and evaluation of FinBERT models on healthcare data, integrating outputs into dashboards and monitoring frameworks, enabling compliance teams to quickly investigate irregular patterns and reducing manual investigation efforts significantly.

• Designed and implemented the Cigna SmartCare Personalization Engine using Python, SQL, and Spark, delivering tailored health plan recommendations that increased policy uptake by 22% and improved member engagement significantly.

• Built scalable ETL and feature engineering pipelines with PySpark, Airflow, and integrated data from MySQL, Hive, and real-time streams, extracting insights from claims, demographics, and member interactions to support data-driven ML solutions.

• Developed machine learning models including XGBoost and Transformer encoders to predict policy lapse, care utilization, and fraud likelihood, improving personalization, supporting risk assessment, and surpassing legacy rule-based systems in accuracy and relevance.

• Applied hyperparameter tuning, cross-validation, and SHAP explainability to ensure model transparency, fairness, and regulatory compliance, achieving Precision@5 81%, AUC 0.91, and F1-score 0.86 for fraud detection with robust performance metrics.

• Deployed containerized ML microservices on Docker, Kubernetes, gRPC, Kafka, and Redis, enabling real-time risk scoring under 100ms latency, maintaining 99.9% uptime, and orchestrating secure pipelines via Azure services including ADLS, ADF, and AKS.

• Collaborated with Product, Risk, and UX teams to conduct A/B tests for AI-driven health nudges, recommendations, and fraud alerts, driving 9% higher policy conversions and 18% reduction in false-positive fraud predictions.