Machine Learning Engineer

What the job is

We are building a decision layer that settles player harm and fraud before anything commercial is allowed to act on the same event. Four tiers, in a fixed order: harm and safer gambling, then fraud and abuse, then compliance, then commercial optimisation. Optimisation only proceeds once the layers above it return clear.

That order is a product decision, and it is also an unusual machine learning brief. The models that matter most here are the ones nobody markets: detecting a player whose behaviour has changed in a way that suggests harm, recognising bonus abuse across accounts, and scoring payment risk in the milliseconds available before a transaction is decided. The recommendation problem is the easy one, and it is deliberately last in the queue.

You would work with our Principal AI and ML Architect and our Head of Machine Learning, on the same platform and the same data that runs the casino, rather than in a separate research track that ships nothing.

The first ninety days

Weeks one to three. Learn the data. Player events, wallet movements, bonus state, the compliance decisions the platform already makes and the audit trail that records them. You will also read the harm and fraud literature we work from, and disagree with some of it.

Weeks four to eight. Ship one model into a shadow deployment. Something narrow with a measurable question behind it, scored against real traffic without acting on it, with the evaluation defined before the model is built.

Weeks nine to thirteen. Take that model to a decision. Either it goes into the pipeline with monitoring, thresholds and a documented failure mode, or you write up why it should not and what you learned. Both are good quarters. Only one of them is a good quarter if you cannot say which.

What we need

• Production machine learning experience, meaning models you have owned after deployment rather than only trained.
• Strong Python, and enough engineering discipline to write code somebody else can run next year.
• Real experience of imbalanced problems and of evaluation that survives contact with reality: precision and recall trade-offs where a false positive has a human cost.
• The ability to explain a model to a compliance officer in plain language, because a decision that affects a player has to be explainable.
• Comfort with the idea that the most accurate model is not always the one we should deploy.

What we do not need

• A PhD.
• Experience in gambling. Fraud, risk, payments, trust and safety, or clinical work all transfer well.
• Large language model experience specifically. Some of this work touches it, most of it does not.
• Enthusiasm for maximising engagement. If the interesting part of this problem for you is making people play more, we are not a good match.

Location and working model

Our engineering base is in Serbia, and this role is either there or remote within European time zones. You would sit inside the engineering organisation rather than in a separate data team, on the same review process and the same on-call expectations as everyone else who ships to production.

Because the subject matter is player harm, this work is done with the compliance team in the room. Expect to spend real time with the people who handle the cases your models are meant to catch.

How to apply

Email contact@scalaralabs.com with the subject line Machine Learning Engineer.

Send a CV or a profile, and a short account of a model you decided not to deploy: what it did, what the evaluation said, and what made you stop.

We read every application. If we are not going to take it further, we will tell you.