Principal AI engineer, eine Rolle, für die Altuon einstellt
Principal AI engineer
The person who can say, in a bank's boardroom, exactly where every byte went.
The principal AI engineer leads the technical side of an artificial-intelligence engagement from the first architecture review to the handover of a running system. The role owns the model and data architecture, the evaluation discipline and the residency boundary, and is the engineer the client's chief information officer calls by name. It is a hands-on role: the principal writes the reference implementation, then makes it possible for others to extend it.
Auf diese Stelle bewerben
- Practice
- Künstliche Intelligenz
- Region
- Alle Regionen
- Beschäftigung
- Full-time employment or a long-term engagement, by agreement.
- Standort
- Remote within Europe, the Middle East or North America, with travel to client sites in the engagement's region.
- Arbeitssprachen
- English; German, French or Arabic in addition where the engagement requires it.
Verantwortung
- 01Design the architecture of each engagement's AI system: where the weights live, where inference runs, how retrieval reaches the client's documents, and how the control plane is kept apart from the data plane. Draw it, defend it in front of the client's risk function, and keep the drawing true as the system changes.
- 02Own evaluation. Build the evaluation set with the client from real material, define what a regression is, run the set on every change and refuse a release that fails it. Evaluation is the practice's product as much as the system is.
- 03Write the reference implementation of each new component and the tests that specify it, then bring the client's engineers to the point where they can change it without you.
- 04Hold the residency boundary. Know the data protection and supervisory obligations of the client's jurisdiction well enough to explain them to an engineer and to a lawyer, and design so that the boundary is enforced by the architecture rather than by a policy document.
- 05Lead a small team: two to four engineers on an engagement, plus the client's own. Set the standard by example, review every change that reaches the client, and make the team's work legible to the engagement director and the steering committee.
- 06Contribute to the practice: the patterns, the evaluation harness, the residency reference architectures and the written positions the firm publishes. What is learned on one engagement becomes the starting point of the next.
Die ersten neunzig Tage
The first month is spent on one live engagement, alongside the practice lead, reading the mandate, the architecture and the evaluation set before changing any of them. The aim is to understand why every decision was made, and to find the two or three that should be revisited.
The second month is the first change owned end to end: a component designed, built, evaluated and released under the engagement's gates, with the client's engineers involved from the design review. By its end the principal is presenting that work to the client's steering committee.
The third month is the first architecture review led alone, on a new mandate or a new phase of an existing one, with the practice lead in the room as a reviewer rather than the author. The residency boundary, the evaluation plan and the exit design are the principal's own by the ninetieth day.
Was geprüft wird
- Architectural judgement
- A design exercise on a realistic mandate with a stated residency boundary and a hostile risk function. We look for decisions with reasons, not a diagram with boxes.
- Evaluation discipline
- A conversation about a system that worked in the demonstration and failed in production. We want to hear how the failure would have been caught before the client saw it.
- Engineering depth
- A review of code you have written and are proud of, and a review of code we have written that has a problem in it. Both conversations are about taste as much as correctness.
- Clarity under scrutiny
- Explaining a technical decision to a non-engineer who is allowed to interrupt. A chief risk officer will do this to you; we do it first.
- Regulatory literacy
- A discussion of what the GDPR, the revised Swiss data protection act or a sector regulator would require of a specific design. Knowing the obligation matters more than citing the article.
Worauf wir achten
- Years of building production systems with machine learning or language models in them, including at least one that a regulated organisation depended on and at least one that you had to take out of service.
- Fluency in the current model landscape and healthy scepticism about it: you can name what a frontier model does better than an open-weight one, and when the open-weight one is the right choice.
- Evidence of teaching: a team, a client's engineers or a public body of writing that made other people better at this work.
- Comfort with the client's room. The role presents to chief information officers, chief risk officers and their auditors, and is judged by how well it survives their questions.
- Working English; German, French or Arabic in addition is valued in the regions where the firm works.
Offene Stellen
Für dieses Profil ist derzeit keine Stelle ausgeschrieben. Bewerbungen darauf werden das ganze Jahr gelesen.
We hire against these profiles as engagements require. Applications are read year-round.