Straight answers on where your data goes, who owns what we build, and what happens when a pilot does not work.
Nowhere. We work inside your cloud account, under your identity provider and your network rules. Nothing is copied to our laptops or to a Lattice tenant, and every engineer signs in with a named account you can revoke the same day.
Whichever passes your evaluation. Every pilot starts with a test set built from your own cases, and we score at least three candidates on it: one hosted, one open-weight and whatever you already license. The contract names no vendor, so the choice can change when the scores do.
You do. Code, prompts, evaluation sets, fine-tuned weights and documentation are assigned to you on delivery, in the statement of work. We keep no licence to reuse them and no copy after handover.
One product owner for about four hours a week, one engineer who knows the data for about eight, and a security contact for the first fortnight. We bring the rest: two engineers, a data lead and a partner who stays on the engagement.
The diagnostic takes two weeks and the pilot eight. Across our last 38 pilots the median from kickoff to production traffic was 11 weeks; the slowest took 19, held up by a data-access review.
You get a written no-go: what we tested, the numbers, why it missed the bar we agreed in week one, and what would have to change. Seven of our last 38 pilots ended that way. The fee is the same, and the test set stays with you.
Your team runs it. Handover includes runbooks, monitoring with alert thresholds and two paired on-call weeks. After that we are off your systems, unless you keep an embedded team by the month.