Senior AI/ML Engineer

I engineered a BERT-based clinical NLP pipeline for multi-label ICD-10 code mapping across 2.3 million unstructured member records. The solution eliminated 14,000 manual review hours and $1.3M in annual rework costs.

I architected a real-time fraud detection ensemble using XGBoost, LightGBM, and Isolation Forest over Apache Kafka event streams. The production system surfaced $8.2M in fraudulent claims within 90 days while sustaining sub-80ms inference latency across 1.4 million daily transactions.

I built an end-to-end MLOps platform with MLflow, automated monitoring, drift detection, and Apache Airflow retraining workflows, reducing deployment cycles from 21 days to 4 days across 18 production models. I also fine-tuned Llama 2 with RLHF for a member-benefits chatbot serving 400,000 daily queries and delivered a LangChain and Pinecone RAG solution for clinical-document retrieval.

I developed a Random Forest churn prediction system with SHAP explainability and automated weekly retraining for a Fortune 500 telecommunications client. The system identified 87,000 at-risk subscribers from a 4.2 million-member base and helped preserve $3.1M in annual contract value.

I delivered a distributed ETL pipeline using Apache Spark and Azure HDInsight to process 6 TB of daily transactional data. This reduced preprocessing latency from 11 hours to 42 minutes and enabled same-day retraining for seven downstream ML pipelines.

I orchestrated ML inference microservices with Docker and Kubernetes, enabling horizontal scaling for 3,500 concurrent API requests. I also implemented self-healing retraining workflows with Grafana, Prometheus, and Apache Airflow, and developed an Azure ML feature-store architecture for standardized, point-in-time-correct feature management.

I built a Python and Django audit risk-scoring engine using supervised-learning anomaly detection for audit analytics. The solution reduced auditor sampling effort by 6,200 hours annually across 14 active client accounts processing 480,000 records per cycle.

I engineered RESTful APIs using FastAPI and PostgreSQL to expose audit analytics to downstream BI platforms. Through query optimization and Redis caching, I improved query response times from 18 seconds to 1.4 seconds for 220 concurrent analysts.

I implemented an NLP-based contract review pipeline using spaCy named-entity recognition and TF-IDF vectorization. The platform streamlined review of 94,000 financial contracts, reducing document review time from 4.2 hours to 38 minutes and producing $940K in annual productivity savings.