Our Mission
Supporting the US private sector in implementing AI in complex organizational processes.
That sentence has been Cerulin’s mission since the first version of this site. What has changed is how much of the work is already half done by the time we arrive.
Where we started
Cerulin began by deploying embedded research teams inside software, financial, and legal operations. The job was to capture the edge-case workflows that never make it into a process document: the exception a senior associate handles from memory, the reconciliation that only works because one person knows which column to ignore, the handoff between two systems that happens over email.
Those edge cases are where automation fails. A workflow that runs correctly on the ninety common cases and quietly breaks on the ten unusual ones is worse than no automation at all, because nobody notices until a client does. Capturing them before anything reached production was the whole point.
What changed
In the last two years the people inside those operations started building for themselves. A partner wires up a workflow over a weekend. An analyst builds a prototype in a chat window that genuinely does the job. The edge-case knowledge we used to have to go and find is now sitting in a prototype someone already made.
And then it stops. The prototype cannot be used for real, because using it for real means putting firm data or client files through a tool that was never built to hold them. The enthusiasm is real, the prototype works, and IT is right to say no.
So the mission is the same, but the starting point moved. We no longer have to discover the workflow. We have to take the version that already exists and make it something the firm is allowed to run.
How we work
Our engineers still work embedded, alongside the person who built the prototype rather than from a brief. That person knows where the prototype cheats, which inputs it was never tested on, and what the reviewer will reject. Rebuilding it without them produces software that is technically correct and practically unusable.
What we hand back is production software: tested, documented, access-controlled, and logged throughout, running on infrastructure that a firm’s security review can actually sign off on. The behavior the prototype proved, engineered to hold up on live data.
What we hold to
-
Sovereign
Firm and client data is processed in the United States, on infrastructure we control, under enterprise agreements with training disabled. It is never pasted into a consumer account.
-
Auditable
Every model call is recorded with its inputs, outputs, and model version, so any result can be traced back and reviewed after the fact. Nothing the software produces is unexplainable.
-
Edge cases first
We build for the ten unusual cases, not the ninety common ones. The common ones were never the problem.
-
Made in America
Engineering, hosting, and support are US based. For firms whose clients ask where their data goes, the answer is short.
We started with professional services firms because they have the sharpest version of this problem: the most client data, the strictest obligations, and the most prototypes sitting unused. If that describes your firm, we would like to see what you built.
Talk to us