Senior Data Scientist - Personalization & Predictions

You'd be joining the Artificial Intelligence team. We own the algorithmic core of the platform: Predictions, Contextual Personalization, contextual bandits, autosegmentation, and the agentic workflows behind Loomi. We work with behavioural data at terabyte scale, across 1,400+ customers, in production, every day. We are currently allowing flexibility for our employees to work from anywhere for the respective region (Central & Eastern Europe) or we are happy to meet you in our offices in Bratislava (Slovakia) or Brno, Prague (Czechia) on a full-time basis.

The mission

You find the signal in how 1,400 brands' customers actually behave — and you prove it moved a business metric. Your models decide which customers are predicted to churn, which segments form themselves, which offer a shopper sees, and whether the discount changed anything or was going to convert anyway.

What you'll actually do

• Frame the problem before modelling it. Most of our highest-impact work arrives as a vague business question, and turning it into something measurable is the first job.
• Work the data at scale. Behavioural data, product catalogues and event streams across BigQuery and Databricks — finding features that carry real predictive signal, not the ones that are easy to compute.
• Build and evaluate models across Predictions and Contextual Personalization: propensity and churn, contextual bandits, autosegmentation, uplift and incrementality.
• Design the evaluation, not just the model. Offline metrics, backtests, and the A/B design that decides whether this ships. You own the question "how would we know if this is worse?"
• Bring methods in from outside and judge them honestly — read the literature, run the quick PoC, and tell a real result from a well-marketed one.
• Hand off cleanly to ML Engineering. You own the model and the evidence; they own making it survive production. That handoff is a document and a conversation, not a notebook over a wall.
• Explain your results to people who aren't data scientists — Product, Engineering leadership, and sometimes customers.

What success looks like after 12 months

• Two models you built are in production and you can name the business metric each one moved.
• A question the team was arguing about is settled, with evidence, and your writeup is what people link to.
• ML Engineering describes your handoffs as easy.

This role bends in three directions

We'd rather shape it around you than the other way round.

• Modelling & data science — features, experiments, and whether the number means anything.
• ML product engineering — you own an ML-powered feature end to end and the ML part is what makes it interesting.
• Platform & MLOps — pipelines, serving, deployment, observability, inference cost.

Most people lean one way and dip into the others. This opening is centered on the first. If you'd rather own the serving path, latency and reliability, our our Senior AI/ML Engineer opening (in Czechia or Slovakia) is the better fit, and applying to both is fine.

What you'll need

• 5+ years building ML models that shipped and were used by someone other than you — in industry, not only in research or coursework.
• Strong Python and genuinely strong SQL. You should be comfortable being handed a warehouse and finding your own way around it.
• Solid grounding in classical ML — tree-based models, regression, classification, clustering — and the judgement to know when the simpler model is the right answer.
• Real rigour on experiment design and evaluation. You know why a metric moved, and when it didn't move for the reason everyone assumes.
• Comfort on a cloud data platform — we work in GCP (BigQuery) and Databricks, but the principles transfer.
• A quantitative degree, or equivalent practical depth.
• Working English, written and spoken.

Plus real depth in at least one of:

• Causal inference and uplift modelling — separating who converts because of an intervention from who would have converted anyway.
• Contextual bandits, sequential decision-making, or off-policy evaluation.
• Ranking, recommendation or information retrieval.
• LLM evaluation, or applied work with the GenAI stack.

Also good: you use agentic coding tools daily and have a view on where they help and where they quietly don't. We build agents for a living, and people who use them tend to have better instincts about them.

#LI-KP1