At World Bank, I develop agent orchestration frameworks using MCP, LangGraph, and retrieval pipelines for secure enterprise workflows. I also built GraphRAG on Neo4j and improved answer accuracy by 31% across enterprise knowledge discovery workloads.
I optimize LLM inference with vLLM, a custom KV cache eviction policy, CUDA kernels, and a C++ PyTorch extension. My work also includes production AI services on AWS and observability frameworks that cut end-to-end latency by 38%.
Previously, at CVS Health, I built distributed LLM evaluation pipelines and automated hallucination detection systems for language model validation. Earlier, I developed machine learning and data systems at Lewis University and JP Morgan, including fraud detection pipelines processing 2B+ transactions annually.
