United States
11 years building AI/ML systems that run in production — not demos. My work sits at the intersection of Generative AI, Agentic Systems, and real-world constraints: HIPAA compliance, enterprise scale, and zero tolerance for hallucinations that affect patients or financial decisions. At United Health Group, I built a PHI-compliant agentic AI chatbot handling prescription inquiries and clinical workflows across digital and voice channels — using LangChain, LangGraph, Azure AI Search, GPT-4, and Claude. The system reduced unsupported responses by 40% via RAG and cut MTTR by 40% through OpenTelemetry-based observability. Inference cost came down 24% by routing tasks intelligently between lightweight models and LLMs. Before that, at US Bank, I built intelligent document processing that turned 1,000-page mortgage packages into structured underwriting outputs in under 5 minutes using Amazon Textract, SageMaker, and DistilBERT/RoBERTa — reducing manual reviewer workload by ~25%. Earlier at Walmart, I built ETA prediction and delay-risk classification for supply chain at DC scale using XGBoost and LightGBM on Databricks and AWS. At Wells Fargo, I built credit risk and fraud detection models that supported regulatory submissions. What I bring to a team: → Agentic AI with safe human handoff, intent routing, and tool invocation → RAG pipelines with vector search (Azure AI Search, FAISS, Pinecone) that reduce hallucinations → LLMOps and observability: OpenTelemetry, Azure Monitor, MLflow, PromptFlow → Multi-cloud ML: Azure, AWS (SageMaker, Bedrock), GCP (Vertex AI) → PHI/PII compliance, HIPAA controls, and audit-ready data pipelines Open to Senior AI/ML Engineer, Staff ML Engineer, and Generative AI Lead roles in healthcare AI, fintech, or enterprise SaaS. 📧 [email protected]
I built an intelligent document processing pipeline that converted thousand-page mortgage packages into structured underwriting outputs in under 5 minutes using Amazon Textract, SageMaker, and Hugging Face Transformers. Confidence-based routing cut manual review workload by ~20%, downstream reviewer touches by ~25%, and CloudWatch/X-Ray monitoring reduced issue detection from hours to under 30 minutes.
I built ETA and delay-risk prediction for inbound supply chain using XGBoost and LightGBM on Databricks and AWS, giving DC teams visibility into late shipments hours before dock arrival. Pipelines ran on PySpark with Delta tables; outputs surfaced in Tableau dashboards and Redshift, with full MLflow model lifecycle management.
I built credit risk scoring and fraud detection models using scikit-learn and gradient-boosted techniques across banking portfolios. Partnered with Model Risk Management on validation workflows that shortened regulatory review cycles, and automated compliance reporting that reduced turnaround from days to standard windows.