I'm an AI Architect and hands-on Principal Engineer with 14+ years across distributed systems, data platforms, and AI. At Autodesk, I lead a team of 4 building the internal platform teams use to create and run AI agents and MCP servers, and I partner closely with our Distinguished Architect on its architecture. We've shipped 3 production agents, 5 MCP servers, and 2 IDE plugins used across the company.
Agents that hold up in production. I design the system around the model — runtime, orchestration, tool access, and deterministic controls — because that is what decides whether AI can be trusted at scale. Proof: an agent I built migrated 1,500+ internal applications off vulnerability-prone APIs to GraphQL across five languages, with over 85% success. The AI does the work; deterministic checks decide what ships.
AI you can measure and operate. Evals for output quality, APM (Dynatrace) for runtime health, and security built into developers' tools, from secure-coding checks to threat modeling.
A better way for engineers to work with AI. I set how my team builds with AI: one agentic SDLC — intent, design, ADRs, specs, implementation, docs — packaged as shared skills and agents, so quality doesn't depend on who is prompting. I coach engineers on when to move fast with AI and when to slow down and design, and I hold AI-generated code to the same bar as any other.
Data foundations underneath. Before AI, I spent a decade building data platforms: Databook's first semantic layer, search at ~200K events/sec, and pipelines over ~2 PB. Agents are only as good as the context they are grounded in.
I still write production code, and I'm most useful where a team needs to move from "we're experimenting with AI" to "AI is part of how we ship."
