Build it where it has to run. Then hand over the keys.
Cohorte is an applied AI company for serious places: banks, hospitals, governments, teams that cannot afford plausible bullshit. We ship production AI systems inside those environments and leave the client's people able to run them without us. Forward Deployed Engineers are how the building gets done. This site explains the role in the open, and gives active FDEs the runbook they work from.
Three doors.
What an FDE does
Scope with the Engagement Lead. Build inside the client's environment. Transfer the system to their operators. Build-and-transfer, never build-and-leave.
The barWho we look for
Shipped production AI into high-stakes environments, with names you can point to. Senior. The things we screen for, and the ones we turn away.
Next stepHow to apply
One application. We review by hand, we reply, and access to the runbook follows. The roster is small on purpose.
What this site is
Two layers, one site.
The public pages (the role, who we look for, the engagement and pay, how to apply, the FAQ) explain what forward-deployed engineering at Cohorte means. Read these before you apply. Anyone can.
The runbook (how an engagement runs, scoping, building in the client's environment, verification and governance, the transfer, security, confidentiality and IP, the reference stack, onboarding, invoicing) is the operating manual for active FDEs. Those pages open once your access is granted, after a short review. They are marked with a small lock in the sidebar.
Why a review. The build roster is deliberately small and hand-picked. FDEs ship code inside banks, hospitals, and governments, where a mistake is not a bug ticket. The review is a short call with the Engagement Lead and one reference. It exists to protect the client and the system, not to gatekeep for its own sake. People who have shipped real systems into real stakes get through.
How it works, in one paragraph
An engagement is scoped by the Engagement Lead, who owns the client relationship and the architecture. You build the system inside the client's environment: their cloud, their data boundaries, their compliance constraints. Nothing leaves. You verify it with the Cohorte method, the same verification and governance discipline we teach, so the system knows when to say "I don't know." Then you transfer it: a Principal Mentor and the client's own operators take ownership, and the system does not leave with us. You are not a consultant who builds and disappears, and you are not handing over a demo. You ship something that holds up in production, and you make someone else able to run it.
The shape of the engagement
Full detail on the engagement and pay page. The honest version: this is senior, high-trust work, scoped tightly, paid as a fixed fee, with a clear definition of done and a real handover at the end.
Who you would build with
Every engagement is scoped by the Engagement Lead. Today that is Charafeddine Mouzouni, Cohorte's founder: an AI scientist who has shipped 60+ production AI systems, built PwC's AI Factory, and published the verification research you would use here (conformal prediction, self-consistency, and TrustGate). You would not be selling or scoping. You would be building alongside someone who has shipped into the same stakes you have.
The reference stack is open, so you can audit it before you ever talk to us: github.com/Cohorte-ai, including TrustGate, the verification work itself. More on the company at cohorte.co, and the research behind the method at teams.cohorte.co/research.
What that looks like in practice: a European bank, six weeks, an analyst assistant that refuses to fabricate a citation, shipped inside their environment and handed to their own team to run.
Think you are a fit?
Read the role and the bar first. Then send one application. No funnel, no sales call, no deck.