Leads the claims adjudication deployment at a European insurer — 3,400 claims a quarter, inside a 14-day regulatory window.
Forward Deployed Engineers, sent into the operation, on a platform built for the job.
Not a list of requirements. A portrait of the engineer who does well here — read it and you will know within a minute whether this is your kind of work.
Leads the claims adjudication deployment at a European insurer — 3,400 claims a quarter, inside a 14-day regulatory window.
Owns the evaluation layer in the Runtime, and the guardrail set every regulated deployment gates against.
Second deployment in dispatch operations, this time handing the on-call rotation to the client's own team in week nine.
A deployment is two to four Forward Deployed Engineers with a direct line to the people whose work changes. They write the code, run the evals, sit in the operations review, and stay until the system is boring.
Every deployment runs on the same substrate: an agent surface your teams use, an evaluation layer that gates releases, and a trace of every decision the system made.
You will not be building the platform from scratch. It exists, and you will extend it.
148 of 3,412 claims breached the window. Three causes account for 91% of them, all upstream of the adjudication model.
A failing gate blocks the release. The engineer who wrote the eval is the engineer on call for it.
Every answer the system gave is reproducible from its trace — the requirement that makes regulated deployments possible at all.
Taught by FDE leads from live deployments, not by a training company. Twenty-four seats, in person, and the curriculum changes as the work does.
Write the test the business would accept, then choose a design that passes it. In that order, every time.
Why most internal agents are abandoned in week three, and the interface decisions that prevent it.
Refusal, scope, citation and audit trail — built as engineering constraints rather than prompt instructions.
Getting a client team to own the pager, and what has to be true before you can leave.
The masterclass is also how some of our engineers found us. Four of the corps joined after attending a cohort.
Two engineers, two weeks, inside one workflow. You get a written assessment of whether an agent changes the number, what it would take, and what would have to be true first.
The role is not a consultant with a laptop. A week in the life of one of ours.
A regulated deployment is reproducible or it is not a deployment. What we instrument first.
Why we stopped writing evaluation sets on behalf of the people who own the workflow.