Selected work
Built, shipped, and still running.
Every one of these ran in a real environment: regulated data, a security review, a budget, and users who notice the moment it breaks. None of it is a prototype somebody parked after the demo. And almost none of it reached production through me alone. Five architects, writing the standards a much larger engineering organization had to follow. Five platform engineers I led and ran the sprints for. Front-end developers I interviewed, hired, and paired with. I still write the hard parts myself.
Flagship
An enterprise multi-agent code-generation platform
Weeks of build work, done in hours
The challenge. A Fortune-500 financial institution was rebuilding the same applications by hand, again and again, against a large proprietary design system and component library. Every screen meant re-reading the standards, re-implementing components someone had already written, and waiting on review. The constraint was never talent. It was repetition, and repetition does not scale.
The architecture. A hierarchy of specialized agents. Some are fluent in the firm’s own component libraries. Some read the real Figma designs. Others write the application and test their own output before a person ever opens it. It runs on Anthropic models on AWS Bedrock, with context engineered deliberately rather than dumped into a prompt, which is what keeps both quality and token cost predictable. Leadership watches a live observability map of what the agents are doing, so the system is answerable in the same way an engineering team is.
The outcome. In production inside a regulated institution, and still saving millions a year in developer cost. I’m a co-inventor on a filed patent for the architecture. It didn’t get there on my own. The design had to go through the teams that own the firm’s design system and component libraries, and through security review, rather than around either of them. Then came adoption, the harder half — engineers learning to trust code they didn’t type. I’ve taken a platform out to more than a thousand engineers before, so I knew going in what that part would cost.
Reusable accelerators — built once, brought to the next engagement
Multi-Agent Systems
Agent hierarchies that plan, act across your tools, and check their own output before a person sees it. The pattern comes out of the platform I co-invented — the domain changes, the architecture holds.
Enterprise RAG & MCP Servers
Knowledge systems grounded in your own data, and MCP servers that connect agents to the business. One in production does speaker-attributed semantic search across millions of meeting-transcript segments.
AI-Native Engineering
Multi-agent code review that reads every pull request, semantic indexing that lets coding agents navigate a large codebase, and agentic sprint operations. Underneath all of it sits the delivery standard: one pipeline I set carried hundreds of applications built on stacks with nothing in common. I set it up, then teach your engineers to run it.
The assessment takes about five minutes and gives you your stage, your real constraint, and the next move worth funding. If you’d rather talk it through, take thirty minutes with me. You’ll get a real answer either way, whether or not we end up working together.