Start with an email
Start with an email
Advising · Training · Speaking

Your agent works in the demo. Now what?

I help teams take non-deterministic systems from "it seemed fine when we tried it" to something they'd put in front of a customer.

Start with an email

Three ways in

no mystery packages, no discovery-call funnel
Advisory

Sit in on the hard part

A standing hour or two with your team while you design, debug, or de-risk an agentic system. I ask the uncomfortable questions early, when they're still cheap to answer.

most useful before you've committed to an architecture

What you get
  • Architecture review of your retrieval or agent design, with the failure modes named out loud.
  • A written point of view you can circulate — not a slide deck, a decision document.
  • Async access between sessions for the "is this normal?" questions.
  • An honest read on whether the thing you're building needs an agent at all.
Shape
Monthly retainer
Who
Teams mid-build
Training & coaching

Evals your team actually keeps

A workshop on your own codebase, or 1:1 coaching for engineers moving into AI work. Everything is built on your data, so it survives the week after I leave.

you leave with a harness in the repo, not a certificate

What you get
  • A working eval harness in your repo, running against real traces.
  • A rubric your team wrote together — the artifact that ends the "was that good?" argument.
  • Hands-on time with retrieval, tool design, and grading, not slides about them.
  • For 1:1: a learning path built around the role you want, plus honest feedback on your work in public.
Shape
Half or full day · or 6 weeks 1:1
Who
Engineering teams & individuals
Speaking

Deep, fast, and funny

Keynotes, conference talks, podcasts, and internal engineering days on agents, retrieval, evals, and doing AI work in Ruby. I aim for the moment where the room goes "oh."

RailsConf, Madison Ruby, Changelog, freeCodeCamp

What you get
  • A talk built for your audience, not a stock deck with your logo on it.
  • Live architecture walkthroughs — real code, real failure modes.
  • Q&A and hallway time; the hallway track is where the good questions live.
  • Workshop add-on if you want the room building, not just listening.
Shape
30–40 min · remote or in person
Who
Conferences & internal all-hands
What it's like to work together
“A rare combination of technical expertise and leadership ability — he took the time to explain new ideas clearly, no matter how complex.”
Jenny Chervenkova AI Strategist, Dynamic Language on building an AI evaluation product together

Is this a fit?

I'd rather tell you now than three weeks in

Great fit

  • You have an AI feature in production or close to it, and reliability is the thing standing in the way.
  • Your team argues about whether output is "good" and nobody can settle it with data.
  • You're choosing between prompting, fine-tuning, and writing actual code — and want someone who's made that call before.
  • You're a Ruby or Rails shop and everyone keeps telling you to rewrite in Python.

Probably not

  • You want a proof of concept built for you end to end — I advise and teach; I don't take on full delivery.
  • You need model training or research work. That's a different specialist.
  • The goal is an AI announcement rather than a working system. I'm not much help there.

How it usually goes

four steps, no theater

  1. You email me

    What you're building, what's misbehaving, and the timeline. Detail helps; polish doesn't.

  2. We talk for 30 minutes

    I ask questions. By the end we both know whether I'm the right person for this.

  3. I send a short plan

    Scope, shape, and price on one page. No proposal deck, no procurement theater.

  4. We start

    First session usually lands within two weeks. You get something usable from session one.

[email protected]