Services
From workflow to production system.
Work takes one of three shapes, set by the decision in front of the team. Discovery decides whether to build at all. Production Spike puts one workflow into production, end to end. Partnership keeps a live system correct once it runs.
The order does not change: map the work as it is, constrain the system before building, ship the smallest production slice, instrument it, and hand over the code, the docs and the runbook.
Partnership is a monthly retainer measured against one promise: no unplanned model migration and no silent eval regression. It covers scheduled eval runs with alert thresholds, successor models tested against the existing eval suite before deprecation, a bounded monthly change budget, named-hours incident response with a stated window, and a written report every quarter. Roughly 10–30 hours a month, banded by system criticality, with a six-month minimum.
Partnership is the one engagement that starts cold. It is bought at handover after a Production Spike, or by a team that inherited a system from another vendor and has nobody on staff who can maintain it.
Typical implementation work: retrieval architecture, model routing, structured outputs, reviewer queues, audit trails, telemetry, prompt and version logging, and integration with identity providers, document stores, LIMS, ERP, CRM or matter-management systems.
Engagements are fixed-scope, quoted against one named workflow, its data owners and a written definition of production. There is no public rate card; the number comes out of a short call.