You buy a defined outcome — our engineers execute it on the pipeline while you watch. Every package is fixed-scope, compute included, priced up-front. No meters, no surprises.
The short version: fixed price, stated baseline, enforced gates, your model. The longer version below.
Because you are buying an outcome — a model that beats a stated baseline — not seat licenses. A fixed-scope package makes the deliverable, the timeline and the price explicit before any compute is spent.
Yes. Every package is quoted with its training compute included. If your run needs more than we estimated, that is our problem, not your invoice.
The pipeline refuses to package it. We iterate within the engagement — different features, more data prep, another template — and if we still cannot beat your baseline, the refund policy applies.
You do. Packaged artifacts (ONNX/safetensors), the model card and the eval report are yours to download and run anywhere. Hosted endpoints are a convenience, not a lock-in.
Production Model includes two re-train cycles. On ML Partner, re-trains run on a schedule you set — weekly or monthly — and each one produces a fresh eval report against the previous version.