Services / Engineering
AI engineering and orchestration consulting.
Senior Staff and Principal engagements on agentic systems, LLM infrastructure, and large-scale distributed backends.
I help engineering teams take production AI work from prototype to operated system. The throughline across my last decade is multi-agent orchestration and the infrastructure underneath it, applied to distributed backends that already carry real traffic. I write code, I review code, I sit in the design discussions that decide which problems are worth carrying forward.
What I work on
Agentic systems and orchestration. Multi-agent architectures framed as selection, expansion, simulation, and a backup phase most production systems omit. The Critic loop, with teeth.
LLM infrastructure. Inference routing, prompt and tool versioning, evals you can run on every PR, model swap without a rewrite, cost and latency budgets enforced rather than aspired to.
Production AI workflows. Job queues, retries, idempotency, tracing across model and tool calls. The unglamorous half of agent systems, where most outages live.
Large-scale distributed backends. Twenty plus years of Ruby, Java, JVM-adjacent stacks, and the kinds of platform problems that show up when a service goes from one region to many.
Technical leadership and incident response discipline. Staff and Principal level work: design review, mentoring, on-call rotations that are humane, postmortems that change behavior.
How I engage
Advisory. A few hours a week. Design reviews, architecture sessions, hiring and team-shape input. Useful when there is a strong team that needs a senior outside read and someone to call when the on-call goes sideways.
Embedded. Half time to full time, integrated into the team. I code, I review, I carry pages. Useful when there is a hard problem and the team wants a senior contributor rather than a slide deck.
Project-based. Scoped delivery against a written outcome with a written end. Useful when the work is bounded and the team wants someone accountable for shipping it. I write the scope before the contract.
How I work
Plan first, then implement. I write the design before the code and circulate it for review. I bias toward small, reversible changes and explicit tradeoffs in writing. I do not ship work I cannot operate.
I keep the same on-call discipline I learned on the engine: show up early, brief the team, name the unknowns, do not freelance under pressure. Incident response is a skill; production AI systems need it.
Representative outcomes
Work that has shipped, framed for a peer.
- Production agentic orchestration at Zendesk: multi-agent assembly of structured outputs from raw inputs, with a Critic revision loop, behind a stable interface that the rest of the product code did not have to learn.
- Distributed backend platforms at Best Buy and Wells Fargo: large-scale services with the unsexy operational properties (idempotency, retries, tracing, capacity planning) that keep them up.
- Hands-on technical leadership: design reviews, mentoring, hiring calibration, and on-call rotations that engineers stayed in.
Not the right fit
What I am honest about not taking on.
- Pure prompt engineering with no production target. If the work ends at a demo, I am not the right hire.
- Greenfield startup CTO with fundraising as the primary deliverable. I build, I do not pitch.
- Pure managerial roles with no code review or design contribution. I lead by writing, not by status meetings.
- Single-shot RAG over a static corpus presented as an AI strategy. Happy to advise on the underlying retrieval; not the product I want to build for you.