AI & Optimisation Engineer

Vacaturebank; Den Haag; On-site; EUR 62,000 - 78,000; Full time

Benefits offered by this job

Electric lease car

Mobility allowance

MacBook or Windows laptop

Claude Code premium seat

25 vacation days

8% holiday pay

Job summary

Lionsville seeks an AI & Optimisation Engineer to design the intelligence layer powering live operations in maritime, logistics and critical infrastructure. You will translate domain knowledge into a semantic model, build knowledge graphs, and craft optimization-driven decision logic with measurable outcomes.

You will combine machine learning with robust software architecture, defend models, and collaborate with senior teams. The role is based in The Hague, with occasional travel to client sites.

Qualifications

• Ownership mindset with clear communication to non-technical stakeholders.
• Experience working with AI in production or live operations.
• Semantic and relationship modelling—ontology and knowledge graphs.
• Expertise in optimization, scheduling, routing, or constraint-based decision logic.
• Applied ML with evaluation, baselines and deployment considerations.

Responsibilities

• Own the intelligence layer of platforms that recommend actions in live operations.
• Model relationships across heterogeneous systems to enable fast, traversable queries.
• Build and maintain the decision logic for scheduling, allocation and routing.
• Apply ML where appropriate, and justify its use or avoidance.

Skills

Ownership

AI knowledge

Semantic modelling

Knowledge graphs

Optimization / OR

Applied machine learning

Software architecture

Communication

Travel readiness

Education

Master's in CS/OR/Applied Math/Physics

Job description

Let your career roar as an AI & Optimisation Engineer at Lionsville

Most systems know facts. Very few know what those facts mean together. This is the seat where a domain becomes a model, a model becomes a decision, and the decision holds up when somebody has to act on it. If the part of the work you enjoy is turning messy operational reality into something a machine can reason over - and then defending it - this is it.

What will you do?

You own the intelligence layer of platforms that recommend action inside live operations - maritime, logistics, industry, critical infrastructure. The vocabulary of the domain, the relationships between its entities, the models that predict, and the logic that decides.

You build the semantic layer

Ontology and taxonomy: what the entities of a domain actually are, what they are called, how
classes relate to each other, and which states each of them may legally be in. Every source system carries its own vocabulary and none carries the shared one - you write that down, make it explicit, and make it something both machines and people can be held to. Getting it wrong is not a naming
inconvenience; it is a wrong instruction on somebody's screen.

You model the relationships

A knowledge graph over systems that were never designed to be joined. None of them knows "the delayed vehicle whose load is destined for a dock already blocked by a half-unloaded competitor, with a cut-off in forty minutes" - that fact lives only in the relationships between them. You materialise those relationships, keep them true as the operation moves underneath you, and make them traversable fast enough to be worth asking during a live shift.

You own the decision logic and the optimisation

A solver that ranks candidate actions against a stated objective: scheduling, allocation and routing problems where the constraints are real and the answer has to arrive before the moment passes. Typically an explicit ordering rather than a set of weights - what is an absolute veto, what is a hard
time-box, what is genuinely being maximised, and what is merely a tie-break. Ordered that way, no setting of any control can trade away the untradeable, and every recommendation can say which constraint was binding.

You put machine learning where it belongs

Forecasting and pattern detection where the data supports them, and rules where it does not - with the judgement to know which. You propagate uncertainty honestly, cap confidence by the weakest input, design the measurement baselines that say whether a model is any good, and retire
the ones that are not earning their place. You will not be asked to chase the state of the art. You will be asked to be right, and to be able to show why.

About Lionsville

We build software that runs real operations - maritime navigation, logistics hubs, critical infrastructure. A business tech boutique: we take on enterprise-critical problems from strategy through to execution, in small senior teams close to the customer. We work as a pride: we back
each other, we say what we think, and we finish what we start.

What you bring to the pride

• Ownership, a bias to solutions, and clear communication - including the ability to explain a model to someone who does not write code, because the people who have to trust it mostly do not
• Experience working agentically with AI - not curiosity about it, but a way of working you already have, and opinions about where it breaks down
• Semantic and relationship modelling - ontology, taxonomy, knowledge graphs and reasoning over them, with the domain modelling underneath: invariants and state machines, and data models whose intricacy is the job rather than an obstacle to it. Which formalism you
  have used matters far less than having built one and lived with it afterwards
• Optimisation or constraint-based decision logic - scheduling, allocation, routing, operations research; lexicographic ordering and hard vetoes as tools rather than jargon. A Master's in computer science, operations research, applied mathematics or physics is the usual route in, or a track record that makes the question moot
• Applied machine learning, and the judgement to know when it is the wrong tool. You can build or consume a forecast, reason about its error, and design a baseline - and we would rather you had shipped something boring that worked
• Strong software architecture capability. The intelligence layer is the spine of the system, so it has to be sound as architecture and not only as a model: clear boundaries, explicit invariants, and enough general engineering that it ships as production code rather than a notebook
• Intellectual honesty about uncertainty, and practical explainability. Saying "the data does not support that claim yet" while everyone would prefer a number, and making sure every recommendation carries its rationale, its confidence and its model version - somebody has
  to trust it at three in the morning
• Living in the Randstad and comfortable with occasional travel to customer sites, including seeing the operation you are modelling at full tilt. The work runs on English; Dutch is welcome.

What we offer

• A gross monthly salary between €5.600,- and €7.000.-
• Electric lease car or mobility allowance
• MacBook or Windows laptop
• Claude Code premium seat
• 25 vacation days and 8% holiday pay per year
• €4.000,- training budget per year
• An inspiring workplace at the Titaan (The Hague)
• Fun outings ranging from culinary delights, social and sporty activities to a family day