Applied Research Scientist (Fraud & Foundation Models) - Sardine

Sardine is the leading agentic risk platform for fighting financial crime. Our integrated solution unifies data across risk teams to help organizations stop fraud in real time, prevent AI-driven attacks, and automate fraud and AML operations. Sardine’s platform is strengthened by one of the fastest-growing fraud consortiums in the market, spanning more than 6 billion profiled devices, 800 million consumers, and 3 million businesses worldwide. Leading companies including FIS, GoDaddy, Intuit, Edward Jones, ZoomInfo, and Checkout.com rely on Sardine to secure and grow trust in their products.

Sardine sits on one of the richest behavioral datasets in fraud and risk: device intelligence, behavior biometrics, session telemetry, payment events, and consortium signals that we leverage to fight fraud across hundreds of fintechs and banks. We are looking for an applied research scientist that brings their expertise in deep learning and foundation models to take this to the next level.

We are looking for an experienced ML applied scientist that can combine foundation model expertise with rich non-text sequential data to come up with practical, state-of-the-art fraud detection solutions. You will have an opportunity to scope and drive the next generation of fraud foundation models at Sardine, and drive industry-wide adoption.

What you'll be doing

• Identify and scope opportunities, design rigorous experiments, and execute on the roadmap for foundation model research and development.
• Own the evaluation bar for foundation model performance: offline benchmarks, time- and entity-aware holdouts, calibration, drift and degradation monitoring, and honest head-to-head comparisons against strong classical baselines.
• Take models the full distance from data prep and tokenization through pretraining, fine-tuning, distillation, quantization, and deployment behind a real-time inference path with tight latency budgets.
• Partner with Engineering on training infrastructure, GPU efficiency, feature and embedding stores, and serving at production scale.
• Work directly with client-facing teams and customers to turn model capabilities and limits into decisions their risk teams can act on.
• Partner with Legal, Compliance, and customer model risk teams to build the explainability, documentation, and governance our bank and fintech customers need to satisfy their own regulators.

What you'll need

• 4+ years in applied machine learning, quantitative modeling, or ML engineering, including at least one foundation model you pre-trained or substantially adapted and put in front of real traffic.
• Hands-on self-supervised pre-training experience, plus practical fine-tuning and adaptation.
• Production experience with model serving, versioning, monitoring, and rollback.
• Ability to self-manage and drive ambiguous applied research projects with clear communication with partner teams across data science, engineering, product, marketing, and external partners.
• Strong Python, strong SQL, and comfort preparing very large datasets.

Nice to haves

• Background in fraud, AML, payments, credit, or adversarial machine learning.
• Experience building and evaluating LLM-based agents in production.
• Publications, released models, or open-source contributions in representation learning or sequence modeling.
• Experience with model risk management and documentation in a regulated financial environment.