Approach
Three engagements, each priced and scoped before it starts. You can stop after any of them — and each leaves you with something your team owns: a decision, a platform, or a measured result.
Which open-weight model should you trust for a defined workload? We benchmark the candidates on your task and hand back a written recommendation, the harness code, and a cost projection.
We deploy a retrieval platform inside your own cloud account — Terraform, dashboards, runbook, eval harness — and train your team to operate it. It runs where your data already lives.
A real document process — loan files, underwriting, claims, contracts, intake — built end to end, fine-tuned where it pays off, and measured against your current cycle time and error rate.
Why it's sequenced
An evaluation tells you whether self-hosting is even the right call before you spend on a platform. A platform gives the workflow pilot somewhere to run. We'd rather sell you the $60K evaluation and stop there than sell you a deployment you didn't need.
Most teams aren't. A 30-minute fit check is usually enough to tell.