Senior Data Scientist - Fraud Model Validation

Klarna; Milano; Ibrido; EUR 90.000 - 130.000; Tempo pieno

Descrizione del lavoro

Klarna is seeking a second-line validator to assess fraud models and governance across real-time payments and identity checks. You will review large-scale datasets, challenge model choices, and ensure production readiness with CI/CD and cloud tools.

You will work with first-line data scientists and ML engineers, applying governance, bias, privacy, and explainability considerations to scale risk management across Klarna's operations.

Competenze

• 3+ years hands-on in fraud-related modeling (transaction fraud, account takeover, identity, or payments fraud).
• Experience with tree-based models like LightGBM and anomaly/graph modeling.
• End-to-end ML lifecycle experience from feature engineering to production monitoring.
• Fluent Python and SQL; scalable data processing with PySpark.
• Ability to design agentic AI workflows and document model risks clearly.

Mansioni

• Assess model performance using fraud-specific metrics and weigh business trade-offs.
• Review transaction datasets and feature pipelines for representativeness and leakage risk.
• Evaluate drift detection, retraining strategies, and production monitoring practices.
• Assess CI/CD and deployment controls - Docker, Jenkins, and AWS SageMaker/Athena/Lambda environments.
• Evaluate model governance documentation, explainability approaches, and regulatory compliance.
• Validate emerging techniques like graph networks and GenAI-based components.
• Document validation outcomes and communicate model risks to data scientists, ML engineers, and stakeholders.

Conoscenze

LightGBM

Tree models

Python

SQL

PySpark

Spark

Model validation

Explainability

AI governance

Docker

Jenkins

SageMaker

Data quality

Formazione

Master's or PhD in quantitative field

Strumenti

Docker

Jenkins

AWS SageMaker

S3

Athena

Lambda

Descrizione del lavoro

Klarna, brieflyAt Klarna, we're building an everyday finance network, helping over 120 million consumers across 26 countries save time and money, and worry less about their finances. Working here means taking on problems most companies never get to solve, and being hands-on enough that the interesting part of the work lands with you, not someone else - you'll build with AI, not watch it happen.

This is the stretch zone. Come find out what you're capable of.

About The Role

First-line fraud teams at Klarna build models against real-time attacks on payments, logins, and identity - trained on transaction volumes north of 100 million records and pipelines with hundreds of features. Your job is to make sure those models actually hold up: independently reproducing results, building challenger models, and stress-testing every assumption from data pipeline to production deployment before a model earns trust at scale.

This is a second-line position, reviewing methodologies built with scikit-learn, LightGBM, graph models, anomaly detection, and increasingly GenAI-based components. You'll also build your own tooling - agentic AI systems that read model documentation and code and surface risks automatically, so validation keeps pace with how fast first-line teams ship.

The scope spans the full model lifecycle: data integrity and feature engineering, conceptual soundness, deployment design across Docker, Jenkins, and AWS, and the monitoring and drift detection that keeps a model honest after launch.

What You'll Do

• You'll assess model performance using fraud-specific metrics - precision/recall, ROC-AUC, PR-AUC, cost-sensitive metrics, and fraud capture rate - and weigh each against its real business trade-off.
• You'll review transaction datasets exceeding 100 million records and feature pipelines with hundreds of features for representativeness, leakage risk, and bias.
• You'll evaluate drift detection, retraining strategies, and production monitoring practices to confirm they catch degradation before it costs the business.
• You'll assess CI/CD and deployment controls - Docker, Jenkins, and the AWS SageMaker, S3, Athena, and Lambda environments models run in.
• You'll evaluate model governance documentation, explainability approaches, and compliance with regulatory expectations on model risk, fairness, and data privacy.
• You'll validate emerging techniques as first-line teams adopt them - graph networks, behavioral biometrics, anomaly detection, and GenAI-based systems.
• You'll document validation outcomes and communicate model risks directly to first-line data scientists, ML engineers, and business stakeholders.

Who You Are

• You've spent 3+ years hands-on in fraud-related modeling - transaction fraud, account takeover, identity fraud, or payments fraud.
• You know tree-based models like LightGBM, anomaly detection techniques, and graph or network models well enough to challenge someone else's implementation choices, not just build your own.
• You've worked across the full ML lifecycle - from feature engineering through production deployment and monitoring - and know where each stage tends to go wrong.
• You're fluent in Python and SQL, and you've used PySpark or Spark to process data at scale.
• You've built agentic AI workflows - not just used off-the-shelf tools, but designed the automation yourself.
• You understand model validation principles and model risk governance well enough to assess bias, fairness, explainability, and privacy risk, not just accuracy.
• You can take a complex model apart, explain what's wrong with it, and make that case clearly to both technical teams and senior stakeholders who aren't.

Bonus points for

• An advanced degree (Master's or PhD) in a quantitative field such as data science, statistics, mathematics, computer science, physics, or engineering.
• Experience in BNPL, credit cards, or other transaction-heavy payment products.
• You've mentored junior validators or led validation reviews.
• Exposure to inference on rejected transactions and how fraud risk and credit risk overlap.
• Familiarity with AI governance frameworks and emerging AI regulatory requirements.

Things you should know before applying

• Working together: we value co-located teams; most teams currently meet in the office 2-3 days per week, and this varies by team and can change over time.
• Non-obvious backgrounds are welcome. Diversity of skills, perspectives, and backgrounds is how we create, innovate, and disrupt like no other.
• Final compensation will be based on the candidate's qualifications, skills, and experience.
• This is a second-line, independent validation position - you'll work closely with first-line fraud data science and ML engineering teams, without reporting into them.