AI Payment Fraud Detection Specialist
An AI Payment Fraud Detection Specialist designs, deploys, and continuously refines machine learning systems that identify and pre…
Skill Guide
A technique that models relationships between entities (e.g., users, transactions, devices) as a graph and applies graph neural networks (GNNs), link analysis, and community detection to identify suspicious patterns indicative of fraud, while using entity resolution to unify identities across disparate data sources.
Scenario
You are given a synthetic dataset of 100,000 transactions between 10,000 users. A coordinated fraud ring uses multiple stolen cards linked to a single shipping address. Build a model to flag the fraud ring.
Scenario
You have a streaming log of financial transfers. The goal is to detect layered money laundering (structuring, smurfing) in near-real-time using community detection.
Scenario
Design a scalable, low-latency graph-based fraud detection system for a global payment processor handling millions of transactions daily, with requirements for sub-100ms inference and model explainability.
Used for storing, querying, and performing native graph traversals (link analysis, shortest path) on the entity-relationship data. Essential for initial exploration and rule-based pattern matching.
Core libraries for implementing, training, and deploying GNN models. PyTorch Geometric is particularly strong for research and prototyping due to its intuitive API and large model zoo.
Spark is used for large-scale data preprocessing and feature engineering. Zingg/Splink are specialized probabilistic record linkage libraries for building the entity resolution pipeline that unifies data before graph construction.
Algorithms for identifying densely connected clusters (potential fraud rings) in large graphs. They are used as a pre-filtering step or as a feature generator for the GNN model.
Answer Strategy
Demonstrate a systematic, pipeline-oriented mindset. Emphasize that the graph model is only as good as the data and entity unification. The answer must start with data sourcing, entity resolution, and graph schema design.
Answer Strategy
Test for problem diagnosis and debugging in a live ML system. Show a methodical approach that separates data issues, model issues, and label issues.
1 career found
Try a different search term.