Skip to main content
Connexion See the demo

Here, a model output ends up signed by a human being.

Our answers go to public buyers, security officers, investment funds and auditors. They legally commit the person who validates them. That changes how the work feels: you are not shipping a convincing demonstration, you are shipping something that holds when it is challenged, eighteen months later.

Who will not enjoy this

If your pleasure is running the newest model on a clean dataset, you will be bored. Most of the work sits elsewhere: traceability, edge cases, the regulatory vocabulary of one particular country, and what the system does when the answer simply is not in the client's documents.

The profiles we look for

Four families of work.

We do not keep a permanent list of open roles: hiring follows the clients we sign. These four families, however, are durable, and a well aimed speculative application almost always finds a conversation.

Applied AI engineering

Designing agents that cite their source, contradict each other when doubt is warranted, and refuse to answer rather than invent. The work is less about the model than about the layers around it: indexing, retrieval, verification, scoring.

What we look at: whether you can explain why an answer was wrong, not only improve a metric.

GRC and regulatory expertise

This is the expertise that goes into the agents. You have prepared audits, written policies, defended an answer in front of an assessor. You turn that practice into rules the system applies: ISO 27001, NIS2, DORA, GDPR, CSRD, public procurement.

What we look at: can you separate what a text requires from what habit has made customary.

Pre-sales and demonstration

Our main demonstration runs live on the prospect's own documents. That means understanding their business in fifteen minutes and owning what the platform cannot yet do. Sales fiction does not survive this format.

What we look at: your reflex when a live demonstration returns an imperfect result in front of the client.

Deployment and client success

Connecting sources, getting the knowledge base indexed, installing in private cloud or on premise, and supporting the first real questionnaire. That is the moment a client decides whether to keep us, and it plays out over a few weeks.

What we look at: your ability to work with an IT department that has doubts, without going around it.

The process

Four steps, three weeks, an answer either way.

A careful read

We read what you write, not how it is laid out. An application aimed at one of the four families always goes ahead of a generic CV.

A framing conversation

Forty-five minutes to check that the role matches what you are actually looking for. We also use it to say what is hard here.

A real case, no trick

A situation we have met: a question with no supporting document, a contested answer, an ambiguous requirement. No timed exercise out of context.

A reasoned decision

Yes or no, with the reason. We ask our clients to validate every answer with a name; we hold ourselves to the same rule.

Write to us, and name the family.

Subject line: "Application" followed by the family of work. In the body, one thing you delivered that held up over time. That is more useful than a cover letter.

hello@optivalue.ai Who we are