Process
Short engagements. Working software. Clean handoff.
Every engagement follows the same shape: understand the decision, build a working slice, prove what it does, and leave your team holding it. Weeks, not quarters — fixed scope, agreed upfront.
Before
Before first contact
Nothing formal. Read the site, kick the tires on the public work — the embedding-map game, the Atlas — and decide whether this is the kind of rigor you want pointed at your problem. A short note about your team, your data, and the decision you’re stuck on is the whole application.
The first conversation is a conversation, not a pitch. If there’s no fit — wrong problem, wrong timing, wrong firm — you’ll hear that quickly and honestly, with a pointer to what would actually help.
During
The three phases
I
Diagnose
A needs assessment on your data, your models, and your team. You get a plain-spoken read: what's working, what's theater, and what to do first. Sometimes the right answer is “you don't need AI for this” — that's the answer you'll get, along with what would actually move the needle.
II
Prototype
A working slice on real data — not slides. Something your team can touch, measure, and argue with within weeks, not quarters, built against an evaluation harness so you know what it actually does before it grows.
III
Hand off
Your team owns it. Documentation, training, and the reasoning behind every decision, so the work survives after the engagement ends. Built to leave — deliberately.
Real data only
No synthetic demos, no mock results. If it can't run on your data, it doesn't ship.
Honest evals
Metrics you'd bet the roadmap on — held out, reproducible, and reported straight, including the bad news.
Senior only
You work directly with the principal. No bench, no handoffs, no juniors learning on your budget.
No theater
If the answer is “you don't need AI for this,” that's the answer you'll get. The firm's only repeat business is the kind earned honestly.
After
After handoff
The engagement ends when your team can run, evaluate, and extend the work without help — documentation, training, and evaluation playbooks in hand. That’s the actual deliverable; the software is just the proof it works.
If you want a longer arrangement afterward — a second phase, a periodic review, a standing evaluation loop — that’s a new conversation with its own fixed scope. What never happens by default: a retainer that quietly becomes a dependency.
Begin
Tell us what you’re trying to build.
A short note about your team, your data, and the decision you’re stuck on is the best way to start. You’ll hear from the principal, not a sales team.
Start a conversation