Help build ML systems that make financial crime harder to hide

Financial crime is constantly changing. New patterns and behaviours emerge all the time, and the data we work with is complex. Our work helps make financial activity safer and more trustworthy for merchants, customers, and SumUp.

As a Senior Data Science/ML Engineer in the Risk AI Engineering Squad, you will build the production systems that turn machine learning into reliable, explainable transaction-monitoring capabilities. You will work across the full model lifecycle: understanding financial-crime typologies, exploring data, engineering features, training and validating models, deploying them at scale, and monitoring their performance over time.

This role is designed for someone who is strongest on the engineering side of machine learning and wants to keep growing their data-science depth. You do not need to be a traditional data scientist or ML Engineer. We’re looking for someone who enjoys working across both disciplines: building robust, production-ready software while staying close to the data, models, and decisions those systems support.

You will join a cross-functional team within the Risk & Compliance tribe, working closely with AML and Fraud Operations, investigators, Product, and Engineering. Together, we build data products and ML solutions that help Risk teams work smarter, faster, and more effectively — while keeping our controls robust, auditable, and compliant across products and markets.

We actively welcome applications from women and people from underrepresented backgrounds. Diverse perspectives make our team stronger and our systems more robust. If you're motivated by technical depth, real-world impact, and the challenge of making ML work reliably in a high-stakes environment, this role is built for you.

What you’ll do

Build ML systems that work in production

• Own and evolve end-to-end batch training pipelines for transaction-monitoring models.
• Build reliable software around the model lifecycle, including testing, CI/CD, versioning, deployment, monitoring, and rollback.
• Improve the maintainability, observability, and scalability of our model pipelines.
• Partner with platform and software engineers to make model delivery repeatable and safe.

Turn data and domain knowledge into better detection

• Build, maintain, and improve ML models for transaction monitoring, balancing detection quality, operational efficiency, explainability, and regulatory expectations.
• Engineer features that reflect AML and Fraud typologies and suspicious behaviours.
• Work with Risk investigators to translate domain knowledge into useful signals, alerting logic, and calibrated thresholds.
• Analyse the drivers of the AML Risk Score and recommend improvements to its features, logic, and thresholds.

Keep models trustworthy over time

• Define and track meaningful model and operational metrics, including detection performance, alert volumes, and investigator outcomes.
• Monitor drift and model health, run back-testing, and investigate changes in performance.
• Run sensitivity tests on synthetic datasets and assess how models behave across relevant scenarios and populations.
• Produce model cards, technical documentation, and other ML governance artefacts that support auditability and regulatory review.
• Contribute to system-design documentation and adapt solutions to regional compliance requirements.

Work across disciplines

• Partner with AML and Fraud Operations, Product, and Engineering to turn ambiguous problems into clear, scalable technical plans.
• Explain trade-offs clearly to both technical and non-technical stakeholders.
• Help the team improve its engineering practices, modelling approach, and understanding of financial-crime risk.
• Share what you learn and support a culture of thoughtful experimentation, constructive challenge, and continuous improvement.

You’ll be great for this role if you have…

Must have

• Strong production Python engineering experience. You write code that ships and are comfortable with automated testing, CI/CD, code review, versioning, observability, and operating services or pipelines in production.
• Experience deploying and operating ML models in production. You understand the practical realities beyond experimentation: reproducible training, model versioning, deployment, monitoring, incident response, and rollback.
• Hands-on experience with end-to-end ML pipelines. You have taken models from data preparation and training through validation and production use, and you understand how to choose appropriate KPIs and evaluation metrics.
• Solid data-engineering fundamentals. You have worked with complex, multi-source data ecosystems and care about data quality, lineage, reproducibility, and failure modes.
• A willingness to deepen your data-science expertise. You are interested in modelling, feature engineering, evaluation, and experimentation — even if your background is primarily in ML engineering or software engineering.
• Clear, confident communication. You can align stakeholders, set expectations, surface risks, and turn ambiguous compliance or operational requirements into a concrete technical plan.

Nice to have

• Experience with PySpark or other distributed data-processing technologies.
• Experience in AML, fraud detection, transaction monitoring, or another financial-crime domain.
• Experience with unsupervised learning, such as anomaly detection or clustering.
• Familiarity with feature stores, model registries, and alerting-threshold calibration.
• Experience producing ML governance artefacts, such as model cards, validation reports, or audit documentation.
• Experience with AI systems and tooling.

This role could be a strong fit if…

• You are an ML engineer who wants to become more involved in modelling and data science.
• You are a software or data engineer who has already shipped ML systems and wants to own more of the model lifecycle.
• You are a data scientist who genuinely enjoys production engineering, automation, testing, and operating models — not only training them in notebooks.
• You like working where technical decisions have a real-world impact and where reliability, explainability, and governance matter as much as model pe