Three stages, and a cheap way to find out
Feasibility review
Two weeks, fixed fee. We look at a sample of your calls and tell you whether this works: which tasks are bounded enough for a small model, which of your languages are viable, what your latency budget actually looks like, where your break-even volume sits.
You get a written assessment either way. If the answer is that you should keep using a hosted API, that is what it will say.
Benchmark and prototype
Six to ten weeks. We fine-tune on your data and benchmark against your current approach on the metrics in our published method. Deployed in your environment so you can test it on live traffic.
You get the model, the evaluation harness and the numbers.
Production
Hardening, integration with your telephony and CRM, monitoring, and handover — or ongoing operation if you would rather not run it. You keep the weights and the infrastructure either way.
What it costs
Engagements are scoped, not listed. The variables that matter are call volume, how many capabilities you want covered, the state of your recordings, which languages you need, and whether you run the infrastructure or we do. The feasibility review is fixed-fee and small enough to be a low-risk way to find out.
Design partner terms
We are taking a small number of design partners at reduced cost in exchange for permission to publish anonymised benchmark results. You get the work at a discount; we get evidence we can point to. What we would want to publish is negotiable and agreed in writing before anything runs.