Senior ML/AI Engineer (m/w/d)

Das ist der Job

What you ship, customers see.

Darum lohnt es sich

The Opportunity QuoIntelligence turns millions of raw signals into finished Cyber Threat Intelligence (CTI) that security teams across Europe act on every day. The ML team is small (2 people today), and the infrastructure is lean.

Quality metrics, ground truth labeling, offline/online evaluation: you’ll design the framework the team uses to measure whether models are working in production. Work directly with the IntelOps team. Your models serve the intelligence operations team; you’ll validate performance against real‑world threat scenarios, not benchmarks.

Python: It’s the team’s language. NVIDIA Triton, vLLM, etc) Familiarity with data orchestration tools (Kestra, Airflow, Prefect) Cybersecurity or threat intelligence domain knowledge (genuinely not required; curious ML engineers ramp fast) What We Offer A small team where your work hits customers.

Your team lead defines priorities; you own how to solve them. The Process Recruiter Screen AI fluency screen Take‑home task: a real evaluation problem from the team’s work. A team member may join. The ML layer is what makes that possible: classification and enrichment today, AI‑powered analysis through Agent Karla next.

You will own end‑to‑end production systems, from improving existing NLP pipelines, fine‑tuning LLMs, building evaluation frameworks, and orchestrating AI agents in the cyber domain. What You’ll Do Improve the production ML stack. The enrichment and classification pipelines work and serve real customers.

They were built for speed, not longevity, so there's room to improve them. You’ll ship at least one measurable improvement in your first 90 days. Own model evaluation end‑to‑end. Ship something the stack can’t do today. The existing pipelines handle classification and enrichment. What comes next is open.

You’ll propose your first project in your first quarter, build it, and measure whether it works. Expand agent capabilities. Help grow Agent Karla’s intelligence by building new orchestration patterns and retrieval strategies using open‑source frameworks like LangGraph.

AI‑First in Engineering You build with AI‑assisted tools daily (Cursor, Claude, whatever makes you faster). But you also know when AI‑generated code introduces risk. You can evaluate whether an AI suggestion is reliable in a security‑critical codebase, and you understand the difference between shipping fast and shipping recklessly.

At a cybersecurity company, that judgment matters more than speed. You evaluate new AI tools critically, adopt what works, and drop what doesn’t. You have opinions on which tools are good and why, grounded in your own usage, not in what you read on LinkedIn. Every model and pipeline has a clear definition of success before it ships.

AI accelerates the iteration loop. Without clear success criteria, that speed is wasted. What You’ll Bring Must‑haves: Production ML deployment. You’ve taken models from notebooks to production and maintained them over time, as part of systems that serve real users. NLP and LLM grounding.

Text classification, NER, summarization, embeddings, transformer‑based models. You understand the fundamentals well enough to choose the right approach for a given problem, not just the newest one. Comfort with messy data. Unstructured text with noisy, inconsistent signals.

If your ML experience is limited to clean benchmark datasets, this role will frustrate you. Open‑source mindset. You’ve worked with Hugging Face, spaCy, OpenNMT, or similar. If your entire career has been inside proprietary ecosystems with no exposure to open‑source equivalents, that’s a blocker. AI fluency.

Active daily use of AI‑assisted development tools. You can show something you built or completed using AI, not just tell us you’re interested.

Nice‑to‑haves: Experience with agent frameworks (LangGraph, LangChain, or similar) and orchestration patterns (ReAct, tool‑calling, multi‑agent systems) Fine‑tuning experience with open‑source models (Qwen, LLaMA, Mistral) Experience with inference servers and popular backends (e.g.

No layers between your model and the intelligence product that clients rely on. Ownership, not just tickets. You’ll review the full service stack and model portfolio, then decide what to change and execute independently. Constraints that force creativity. We believe in using the simplest solution that gets the job done.

We keep things lean and pick tools carefully. The interesting problems here come from making that work: squeezing more out of distilled models and designing pipelines smart enough to ship on the available hardware. Ethical red lines. QI holds the "Cybersecurity Made in Europe" label and serves as ENISA (the EU’s cybersecurity agency) partner.

We’re upfront about what our AI can and can’t do, and our compliance record backs it up. In a market full of AI hype, that credibility is a real advantage. Growth trajectory. Pre‑Series A, revenue growing fast, Series A planning underway. You’d be shaping the ML direction for a company that still has ~40 people.

You can use AI tools, but you’ll defend your reasoning and choices in the next round. Hiring manager interview: walk through the task, then broader technical and behavioral assessment. CEO/CTO Interview Offer and Background Check We welcome applications regardless of gender, nationality, ethnic origin, religion, disability, age, or sexual identity.

Diversity is key to producing high‑quality intelligence. #J-18808-Ljbffr