At Abbott Laboratories, I help build and run a HIPAA-regulated clinical AI platform on AWS Bedrock that handles 50K+ LLM requests a day and has cut clinician documentation work by 35%.
I built an evaluation harness to score groundedness and flag hallucinations, and improved serving performance with response caching and request batching. These changes brought p95 latency down from 3.2 seconds to 0.9 seconds and inference cost down 28% without hurting answer quality.
I designed hybrid retrieval over clinical records and built multi-step Q&A agents with LangGraph, MCP tool calling, and human review. I also added serving-layer guardrails for prompt-injection detection, PII redaction, and output validation.
Previously, at Wells Fargo, I worked on the MLOps platform for real-time fraud detection, improving detection 18% over legacy rules. At CareEdge Ratings, I automated financial ETL and built data pipelines and Snowflake models for regulatory reporting and credit-risk KPIs.
