We're looking for an experienced AI/ML engineer with deep domain expertise in financial crime to join our fintech startup at the ground floor. If you've built fraud models before and want to apply that expertise to the biggest unsolved problem in payments, read on to learn more ⬇️
About the role 🎭
Banks approve or reject payments with almost no context. We're building the intelligence layer that changes that — running real-time investigations on payments before they clear. You'll own the models and machine learning systems that power our fraud prediction: turning messy, multi-source evidence into risk scores that banks trust and act on.
What you'll work on
- Design, build, and iterate on ML models and AI products that underpin our fraud detection. Feature engineering, guardrails, evals, production deployment, monitoring — you’ll work on the full end-to-end
- Develop fraud prediction capabilities that go beyond sender-side signals: scoring recipient risk, payment context, and situational indicators using labelled fraud data and feedback loops
- Improve the capabilities of our investigation engine to turn unstructured, multi-source evidence into features that improve model performance
- Build and maintain the MLOps infrastructure needed to train, evaluate, deploy, and monitor models in production with the reliability banks demand
- Collaborate closely with engineers, product, and our bank customers to understand where model improvements have the highest impact on fraud prevention outcomes
- Bring your domain expertise to bear: help shape our product roadmap by identifying which fraud typologies, signals, and intervention strategies will move the needle most
- Experiment with and deploy LLM-powered approaches where they outperform traditional ML or provide faster time-to-market — we're pragmatic about techniques, not dogmatic
- Contribute to our data strategy: what labels do we need, how do we close feedback loops, and how do we build compounding data advantages over time
This is not a research role. You'll ship models that make real-time decisions on real payments. Engineers here drive product decisions and own systems end-to-end.
Why this is hard
Fraud is adversarial, low-prevalence, and high-stakes. Labels are noisy and delayed. The evidence that distinguishes a scam from a legitimate payment is messy — unstructured, scattered across sources, rarely behind clean APIs (but you know this already!). Models need to be fast enough for real-time payment flows, explainable enough for banks to trust, and robust enough that adversaries can't trivially evade them. And we're critical infrastructure: when we're wrong, real people lose money.
Our stack today:
- Backend: Python, Kotlin
- Infra: Cloud-native, event-driven services, async workflows (GCP managed using Terraform)
- Data: BigQuery, Elastic; correctness > dashboards