Services
From workflow to production system.
There is no catalogue. Work takes one of three shapes: Discovery finds the production-shaped problem, Production Spike ships one workflow end to end, and Partnership keeps a live system correct after launch.
Which one applies depends on the decision in front of the team: whether to build at all, how to launch one workflow safely, or how to operate a system that is already running. The delivery model is intentionally direct: map the real work, constrain the system before building, ship the smallest production slice, instrument it, and transfer the runbook to the client team.
Partnership is a monthly retainer for sustained operation: continuous eval monitoring with alert thresholds, ownership of the model migration path so successor models are tested against the existing eval suite before deprecation forces a swap, a bounded monthly change budget, named-hours incident response with a stated response window, and a quarterly written report for leadership. The measurable promise is no unplanned model migration and no silent eval regression. Roughly 10-30 hours per month, banded by system criticality, six-month minimum.
Partnership has two entry paths. It is bought at handover following a Production Spike, or bought cold by a client who inherited an AI system from another vendor and has nobody on staff who can maintain it. Discovery and Production Spike are not prerequisites for that second path.
Typical implementation work includes retrieval architecture, model routing, structured outputs, reviewer queues, audit trails, telemetry, prompt/version logging, and integration with existing enterprise systems such as identity providers, document stores, LIMS, ERP, CRM, or matter-management tools.
Engagements are fixed-scope and quoted against one named workflow, its data owners, and an explicit production definition. There is no public rate card; scope and fee come out of a short intro call.