How we work
The sequence doesn't change — each step exists to de-risk the next one.
One to two weeks. We map the problem, your data, your constraints, and what "working" means in numbers — and interview the people who do the work today.
It ends with a written brief: approach, risks, feasibility, and a fixed price for the next step. If AI is the wrong tool, it ends with that finding instead.
Two to six weeks. A thin version of the real system on your real data, with the evaluation harness built first — so "is this good enough?" is answered by measurement, not by the best demo.
The prototype answers the go/no-go question cheaply. If the numbers don't clear the bar, you learned that on a fraction of the budget.
Four to twelve weeks. Hardening, observability, cost controls, failure handling, human-review routing, and documentation your team can operate from.
This is where model-agnostic architecture pays off: when a better model ships, switching is a measured configuration change — not a project.
Your team owns everything from day one. We either hand over cleanly, with a walkthrough and transition period, or continue as an engineering partner on a monthly retainer you can end any month.
Pricing
Discovery is a fixed fee; every step after is priced in the previous step's written output. Stop at any boundary and keep everything. No open-ended time-and-materials — it puts our incentive against yours.