Twenty years of production engineering, currently focused on AI systems — LLM-powered apps, RAG pipelines, multi-agent architectures, and the substrate around them. I take work where the deliverable is judgment, not keystrokes — designing what should be built, reviewing what gets built, and owning the architecture under it.
How to engage me
AI architecture, embedded
Three to six months, integrated with one team. Designing the AI system — retrieval, tool use, orchestration, evaluations, MCP integrations, data layer, security boundaries — and reviewing the implementation as it ships. Single LLM in the loop or a full multi-agent topology, whichever the problem actually wants. I direct coding agents for the build; the spec, the review, and the design are mine. Best fit when you're past the prototype and the next version needs to survive contact with production.
Architecture review
One to three weeks, written deliverable. A senior pair of eyes on a system you've already designed, or on one that isn't behaving the way you hoped. Output is a written report: what's load-bearing, what's brittle, what I'd change first, and why. Short engagement, no ongoing relationship implied.
Full-time
Senior AI architecture roles at companies that take engineering quality seriously and put "ship it" as the second sentence, not the first. Open in principle, selective in practice.
How I work
- The spec is the artifact, not the prompt and not the code. A clear specification beats a clever implementation, and it's what the team actually reasons about.
- Evals or it didn't ship. Behavior worth shipping is behavior worth pinning down — I expect a held-out scenario suite for any agent we plan to put in front of real work. My take on why.
- One source of truth per fact. Duplicated state is duplicated bugs. I look for it actively in reviews.
- Boring tech where it can be boring. Postgres, FastAPI, Next.js. Novelty belongs in the AI layer; the substrate should be predictable.
- Async by default. Long writing beats short meetings. Decisions land in documents so the next person doesn't have to rediscover them.
What I don't take on
- Roles where the deliverable is keystrokes rather than judgment — staff augmentation, IC implementation work with a fixed spec. Not because it's bad work, but because it's not where my edge is.
- Engagements with no production target. I'm interested in the systems that have to survive a Monday morning, not the demos that have to survive a Tuesday afternoon.
- Anything that requires me to be the person typing. The whole point of how I work now is that I'm not.
Background
Twenty years across backend systems, distributed data, payments, LLM-powered SaaS, and now production AI architecture. Shipped at small startups and at scale; led teams and contributed as an individual. The throughline is engineering judgment under real-world constraints — full CV on request after a first conversation.
Get in touch
The fastest path is the contact form. Tell me what you're building, what stage you're at, and what specifically you want a second pair of eyes on. I read everything and reply within a few days; long-form is welcome.