CDI - Manager AI Engineer - Paris

Mission New models, open or closed, come out every month: the challenge is still to choose those good enough for production, and to prove it. Applications and agents rely on large language models (LLMs), and the model part of these systems is in your hands. Our clients use them to answer their own customers, process documents and query their knowledge. What the model sees, the way it retrieves knowledge, and the tools it can call: these choices are yours, as is measuring the quality of the answers. The chosen model must meet the cost and quality objectives. This position is made for you if you want to know whether an AI system gives correct answers, and not just whether it runs. Artefact supports the largest companies in the world in deploying AI: 3,000 AI makers in more than 30 countries, more than 1,000 clients. In France, the Engineering department brings together about 300 people under the responsibility of the CTO; you will join the AI Applied Engineer chapter, AI Engineer archetype. Activities and responsibilities Product Engineers build the application around the model; the behavior of the model inside it is your responsibility. Context & Analytics Engineers prepare the knowledge your agents retrieve, and Platform Engineers operate the serving, the harness and the observability you rely on. With the AI Product Manager and the client's business experts, a good answer is defined, and then the results are reviewed together. Peer code review and mentoring of new hires are part of the daily routine. Your responsibilities: - Build LLM applications and agents: context design, tool calling, multi-step workflows. - Design how agents use knowledge: retrieval queries, re-ranking, citations. - Build evaluation sets and pipelines: reference answers, automated scoring, review by business experts. - Monitor the quality of answers in production and fix what degrades. - Choose models based on quality, cost and latency; do fine-tuning or distillation on a smaller model when prompting is not enough. - Add safeguards against wrong answers, data leaks and prompt injection. - Process documents with LLMs: extraction, classification, summaries. Profile sought Essentials: 1. You have built an application or agent that calls an LLM, and you can explain what worked and what failed. 2. Your code is tested and maintainable. 3. Coding agents are already part of your daily work, and you verify what they produce. 4. The quality of AI outputs is measured in your work with test sets and metrics, and you know how to limit hallucinations, prompt injection and data leaks. 5. Putting an LLM application into production and maintaining it (deployment, monitoring, cost, latency) holds no secrets for you. 6. Your starting point is the business problem rather than the model; recognizing that AI is not always the right tool, and explaining this to non-technical people, is part of your practice. A plus, not a prerequisite: - Fine-tuning or distillation of language models. - A training in machine learning or statistics, for example in data science. - Agent frameworks or protocols, such as MCP. - Multimodal inputs: documents, images, voice. - LLM observability tools: tracing, cost tracking. No need to tick every box to apply: if you have most of the essentials and the desire to learn the rest, let's talk. Why join us? In this job, the difficulty is less about making a model talk than about knowing whether it says the right thing. The AI Engineer answers this question through measurement: test sets, evaluations, monitoring in production, rather than the impression of a successful demonstration. We are AI makers: AI is built with our clients, not just for them. Their challenges become ours, and our technical choices are explained in the language of the business. The positions we take are clear; they evolve in the face of facts, never under pressure. Our field changes every month, and this pace suits us: new technologies and new sectors are learned quickly, and habits that no longer serve us are abandoned without regret. Faced with a difficult problem, the reflex is to break it down into verifiable steps, with a high standard of requirements for both code and deliverables. When something is missing, for a client or for the chapter, we build it. Progress comes through constructive feedback, knowledge sharing, asking for help without waiting, and recognizing the work of others. There is a lot to learn, and a lot to build. A counsellor supports your first months. The technical onboarding opens the journey, followed by a first client project; if none is ready, the internal starter kits take over. The rest is deepened on projects: evaluation, agent design, model adaptation, with the support of the AI Engineer community. From 2027, the AI Engineer archetype will have its own skills framework and training path, in addition to the common base shared by all AI Applied Engineers (production responsibility, fundamentals of AI security). Later, a cross-functional role becomes possible (Project Tech Lead, Pre-sales Engineer, Engineering Manager), as do the AI Architect and AI Product Manager chapters. Recruitment process The process takes about four hours, divided into six steps, only one of which is on site. 1. HR interview (45 min, remote): background, motivations and practical points, such as schedule and salary expectations. 2. Call with the CTO (15 to 30 min, remote). 3. Technical screening (30 min, remote, with a member of the AI Applied Engineer chapter): detailed feedback on your past projects, then technical and solution design questions. 4. Code review and live coding (60 min, on site, with a member of the chapter): review of an existing code repository and fixing its defects, then building a new feature with a coding agent and presenting the result. 5. Technical case (45 min, remote, with a member of the chapter): framing a business problem, proposing a technical solution, describing the evaluation method, then discussing your choices. 6. Final interview (30 min, remote, with a member of the department's management, Engineering Manager or Director): revisiting open points, choosing the archetype and level best suited to your profile, and answering all your questions. Depending on the step, all AI tools are allowed, or none. During live coding, the way you work with AI assistance is observed. Compensation: 75,000 € to 85,000 € gross annual (fixed) depending on experience and seniority + up to 8,000 € variable depending on experience and seniority *At Artefact, we believe that the diversity of backgrounds, experiences and perspectives enriches our teams. We are committed to providing an inclusive environment and equal opportunities, where everyone can grow and fully express their potential. Our recruitment is based on skills, potential and fit with the role.* Find Jobs in France on Arbeitnow