About Me

Building Software Since 1999

I'm a Principal Software Engineer and lead the AI Studio team at Leaf Software Solutions, where I architect custom AI solutions for software customers.

I've been building software professionally since 1999. Over that time I've worked across application architecture, distributed systems, cloud infrastructure, DevOps, mobile, developer tooling, and product engineering. I've always been most interested in difficult problems where the answer isn't obvious yet.

Today, many of those problems are in AI.

Beyond Calling an LLM

My work with AI goes considerably deeper than integrating model APIs into applications. I'm interested in the systems around the models: the infrastructure that runs them, the harnesses that turn them into agents, the environments where those agents act, and the evaluations that determine whether they actually work.

I've built agent runtimes and harnesses, a framework for constructing new harnesses, agent libraries in Rust, Python, TypeScript, and Go, multi-agent systems, evaluation frameworks, software engineering benchmarks, isolated execution environments, and autonomous software development systems.

I also operate my own AI compute infrastructure, working directly with model serving, vLLM, inference optimization, quantization, KV caching, long-context workloads, model distribution, LLM proxies, and routing requests based on capability, complexity, latency, and cost.

Autonomous Software Engineering

The question I keep coming back to is what happens when AI moves from being a tool used by a software engineer to becoming an active participant in the engineering process.

Making that useful requires much more than a capable model. Autonomous agents need safe places to execute code, reliable tools, measurable objectives, reproducible evaluations, and ways to independently verify what they actually did.

That has led me to experiment with autonomous coding systems, dark-factory approaches to software development, agent evaluation and verification, and even self-improving coding agents that modify their own implementation and have the next generation independently benchmarked to determine whether they actually improved.

What I Work On

Building Systems We Can Trust

As agents become more capable, I think one of the most important engineering problems is learning how to give them greater autonomy without giving up observability and control.

One example is Petri, an AI sandbox system I built around OS-native microVMs. It provides isolated environments where agents can execute code while maintaining explicit boundaries between the agent and the host.

I've also built Remiss, a tool that audits coding-agent runs by comparing what an agent says it accomplished with its transcript and actual side effects. It turns the run into a reviewable verdict based on what actually happened, rather than simply trusting the agent's report.

Always Building

I've never been particularly good at separating work from hobbies because most of the things people do professionally are things I'd probably experiment with for fun anyway.

My projects have ranged from AI infrastructure and developer tooling to SaaS products, mobile apps, IoT, electronics, CNC, laser engraving, 3D printing, and whatever technology has caught my attention lately.

Learning has always worked best for me the same way: build the thing, push it until it breaks, figure out why, and build the next version better.