At U.S. Bank, I own production reliability for fraud machine learning services on AWS and Kubernetes, monitoring services that support 2M–4M+ daily transactions. I also manage model refresh cycles, improving precision by 8–12% at comparable recall.
I built FastAPI inference services supporting 2K–5K requests per minute and integrated them with Java/Spring Boot microservices. MLOps monitoring and release improvements cut unexpected model-performance degradation by 20% and improved deployment reliability by 30%.
At U.S. Bank, I built LLM-powered investigation workflows and RAG and agentic AI capabilities using GPT-4/GPT-4o, LangChain, LangGraph, and Pinecone. These workflows reduced analyst review effort and improved investigation efficiency by 20%.
At TCS, I delivered production machine learning solutions across healthcare, retail, and telecom engagements. My work included data pipelines processing 5M+ records, forecasting and anomaly detection, and NLP classification pipelines that processed 10K+ documents monthly.
