Agentic AI, in production
Production MCP servers, agent runtimes on Bedrock, and RAG pipelines built to survive contact with real data — not demos.
// What I do
Three places I move the needle
Deep specialization where it counts, backed by two decades of building systems that were meant to last.
Dual-cloud architecture
Azure and AWS as first-class citizens — including the cost modeling, networking, and trade-offs that decide which workload lives where.
Identity & integration
The connective tissue: OAuth2/OIDC, SAML federation, Entra/B2C, and the API layer that binds enterprise systems together.
// Selected work
Things I've actually shipped
A sample of recent builds — the full history lives on the experience page.
01
Two-cloud agentic AI platform
A C# backend exposing MCP HTTP endpoints from Azure, an Angular chat client on AWS, and a Python agent on Bedrock — a production-oriented pattern for running agents across cloud boundaries.
02
RAG system built from the ground up
Embeddings, chunking, retrieval, and evaluation (recall@k, MRR) implemented from scratch before extending to Qdrant — then mapped onto Bedrock Knowledge Bases and Azure AI Search.
03
Serverless medical-billing pipeline
An event-driven AWS architecture (Lambda, DynamoDB, Step Functions, EventBridge) processing billing files for populations from 500 to 1,000,000 people, with Athena/QuickSight reporting.
// Let's talk
Stuck between a pilot and production?
That's exactly the gap I close. Tell me what you're building and where it's stuck.