Availablefor talks, workshops, and advisory
I build and explain the systems underneath AI, Azure, and GitHub, from first principles to production.
Right now that means agent architectures, MCP servers, and the security and governance questions that show up the moment AI tools reach a real engineering team. All of it rests on three decades of software engineering, technical publishing, and live instruction, which is how I learned to explain hard systems to the people who have to run them. I'm a Microsoft MVP, a Microsoft Certified Trainer since 1997, an author for Microsoft Press, Pluralsight, and O'Reilly, a Microsoft Ignite speaker, and an Anthropic Partner.
2M+Watch hrs
300+Courses
1M+Learners
1997MCT since
What I'm building.
My current work sits where AI agents, the Model Context Protocol (MCP), GitHub, and Azure security overlap. Everything here is a public repo you can clone and read, and each card says plainly whether you're looking at a prototype or a teaching reference.
Agent Security Quorum
◈PrototypeAgent and skill definition files are executable attack surface that happens to look like plain Markdown, and no regex is ever going to catch a prompt injection written in natural language.
So every pull request that touches those files goes to four independent evaluators, three LLM lenses on different model families plus a static scanner, and the build fails only when two of them agree. That quorum costs extra model calls on every PR, and in exchange no single model's blind spot can wave a malicious file through.
Trusted MCP Registry
◆Teaching referenceOnce developers can wire any MCP server into VS Code, somebody has to decide which servers the organization trusts, and that decision needs a review trail.
I built it as a policy plane rather than a runtime: a Git-tracked allowlist changes only through pull request review, and a PowerShell script writes each developer's mcp.json from the registry. VS Code doesn't read a remote registry URL, so that sync script is the real enforcement point, and the README says so up front.
Context Engineering with MCP
◆Teaching referenceAI assistants forget everything between sessions, which makes long-term memory an architecture problem rather than a prompting problem.
The repo climbs from a hello-world MCP server to WARNERCO Schematica, a production-shaped app that pairs FastMCP with a LangGraph hybrid-RAG pipeline over vector and graph stores. The labs build in layers, so you understand each moving part before Schematica puts them all together.
The rest of my public work, including more MCP servers, GitHub Advanced Security labs, and the code behind my courses, lives in these two GitHub accounts.
