Ongoing support for production AI systems
We keep your live AI system accurate and current, then use the remaining engineering capacity on the next improvement you have agreed.
What you get
A live system degrades quietly. A policy is reissued and the search still returns the old version. A group membership changes and the index never hears about it.
Keeping up with that is our job, on a fixed engineering capacity each cycle that you agree before it starts. The system stays in your environment on your accounts, your own admin keeps running it, and either side can leave on 30 days' notice.
What you get each cycle
- A report at the end of each cycle
- What shipped, what it scored, and what is queued next.
- Ongoing checks on answer quality
- Quality, speed and cost, traced on every request.
- Re-testing when your documents change
- The evaluation set rerun and scored before release.
- Access changes kept in step
- Permission updates reach the search index, not just the screen.
- Model upgrades tested before they go live
- Each new model scored against the one you run today.
- A search index you can roll back
- Rebuilt when the setup changes, and revertible if one goes wrong.
Questions
The system keeps running. It sits in your environment, on your accounts, and the code and prompts were yours from handover, so ending the retainer ends the engineering capacity rather than the service. Either side gives 30 days' written notice after the six month minimum term.
A cycle is a fixed engineering capacity rather than a ticket allowance, and unused time does not roll forward. Cycles where nothing breaks go to the next item on your shortlist, which is why the close note names what shipped.
The engineers who built the system. There is no account layer and no first line, so the person reading your report is the person who can change the code.
Keep the system worth having
If a workflow is live and nobody has rerun its evaluation set since launch, that is the conversation.