Work with me

I work as an independent principal: one or two substantial engagements at a time, hands-on, end to end. There is no team behind me and no agency in front of me. The person on the call is the person in the repo.

Three ways to work with me

Build · 3–12 months

Build

A system that doesn’t exist yet (an agent pipeline, a knowledge graph, an entity-resolution layer), built and shipped with its tests and its documentation. The Orpheus stack is the reference: four sources reconciled into one schema, a deployed product, an agent auditing it, and an oracle answering questions over it.

You get the system and the decision records.

Advise · a few days a month

Advise

Fractional principal for a team building its first agent infrastructure or untangling heterogeneous data. Architecture reviews, instrumenting LLM systems so you know what they cost, deciding which parts of a problem a model should not be doing at all, and making the dashboards tell the truth.

You get a second opinion you can check, on retainer.

Measure · 2–6 weeks

Measure

You have an AI system answering questions over your own data, and no way to tell whether last week’s change made it better. Most teams never built the test, because the test is the tedious part. I build it, from your real questions, and hand it over:

  • A few hundred of the questions your users actually ask, each labeled with the documents that should answer it.
  • A harness that scores your shipped system, not a reimplementation of it.
  • A baseline, broken out by kind of question, so you know which kinds are weak.
  • A written account of what the score cannot see.
  • All of it yours to re-run on every change from then on.

You get a number you can defend, and the instrument that keeps producing it.

Start here · 1–2 weeks · from $10,000, fixed

The scoping audit

Every engagement can begin with a short, paid scoping audit: I read your system or your data, and you get written findings, a costed plan, and a build / fix / don’t-build recommendation, whether or not we go further. If we do, the fee is credited in full against the engagement. I take on two or three audits a quarter.

How I work

I write a decision record for every choice that matters, start tests on day one, measure costs instead of estimating them, and keep the agent’s reasoning in the log. If that sounds like overhead, we are not a fit. In my experience it is what keeps a project moving in month three.

The obvious worry with hiring one person is what happens when he leaves. Everything I build assumes someone else will maintain it: a project wiki and decision records grow with the code, so your team can pick the work up cold.

A fit if

  • your data is heterogeneous and nobody has reconciled it.
  • your agent system has to be legible, not only good in a demo.
  • your team wants the thinking written down along with the artifact.
  • the right full-time principal role, occasionally.

Not a fit if

  • it is body-shop staff augmentation.
  • the brief is “make the demo look good by Friday.”
  • it is commodity dashboard work.

Start

Write to daedalus@s-martens.com with two paragraphs: what you’re building, and where it’s stuck. I read everything and answer quickly.